skills

[rustic hound]

has there been an explination of why the "Skill flagged — suspicious patterns detected" banner appears even when both Security Scans say "Benign"?

[uncut dome]

got a link to one?

[fresh wing]

Hello folks! I have two installations running, one local and on a server (did not want to open my local machine to telegram etc). In order to sync the workspaces I started experimenting with synching the two workspaces via github. Each installation synchs its changes to branches of the repo. Claude and I built a skill that merges all changes back into main, which the machines then pull to be up to date. Hope I explained that well. Anyhow, if you like the idea let me know. If you don't let me know. Not sure if I am allowed to post my github repo for claw-sync here?

[timid canopy]

Someone gift me Claude pro plan I'm so broke 😭

[blazing blade]

Hello guys, I'm having an issue: In Discord, when replying to a message and mentioning the OpenClaw bot, I want it to automatically create a Linear issue from that replied message, assign it to me, and subscribe anyone else I mention in my reply.
I tried making a custom skill for this, but OpenClaw keeps failing to trigger it. It says the words like 'todo', 'issue', or 'mark' are conflicting with built-in system keywords or AI intent matching, so my skill gets overridden.
How can I fix this?

[rustic hound]

yeah here is an example https://clawhub.ai/kaimcpheeters/skirmish

[crude hinge]

How are you guys finding the built in skills for the Apple ecosystem? Like notes etc

[viral portal]

I haven't used Notes (big Evernote user... probably time to leave that ecosystem).

But the built in support for Reminders is fantastic. It's my go to example. Trying to add reminders with Siri is useless... it's like Siri has no idea who you are or what list you are talking about and trying to remember the verbal syntax to create a list is impossible. But with Claw... "Add an item to the household paint list to paint the downstairs bathroom, if I don't have it there already". Boom... done. Even if the list is called "Painting" or "Household - Paint" or whatever. You can also tell it "move all the paint tasks off the Household list onto a Painting list and it'll do it. I love it!

[wide ember]

any good reference for troubleshooting skills?

[old frost]

skills aren't only an openclaw thing look up anthropic skills they have a huge guide on it

[stuck frost]

if yall have ever had OC forget its doing something or it crashes while doing something and actually want it to complete stuff that you asked it to do
https://clawhub.ai/npmisantosh/task-system

[half mulch]

Just dropping in to leave my 2 cents. recently i started using SearXNG for my searching skill. it runs locally and seems to work really well with my agent

[fair plover]

Hey! we just published TokPortal as an OpenClaw skill on ClawHub.

This skill is for mass social media account creation and video distribution (TikTok, Instagram).
It wraps our API into 30 MCP tools:** create accounts, configure videos, track analytics, upload videos/slideshows**, all from natural language.

Install:
clawhub install tokportal

Then set your API key in ~/.openclaw/openclaw.json and you're good to go.

More info: https://clawhub.ai/naybu256/tokportal

[hallow mural]

I built https://clawtick.com — a cloud scheduler for OpenClaw agents.

OpenClaw's built-in scheduler silently drops jobs. No logs, no errors, just missed messages.

ClawTick fixes that:
→ Reliable cron in the cloud
→ Dashboard + CLI
→ Delivers to WhatsApp, Telegram, Slack, Discord

[fresh grove]

This drove me absolutely mad for weeks. My agent would lose track of what it was doing mid-task, especially around context resets and compaction.

I ended up building a 5-layer observational memory system to fix it. Basically a Gemini Flash agent watches conversations every 15 mins and extracts durable facts to markdown files, plus a reactive file watcher, pre-compaction hooks, and session recovery checks. Belt, braces, and three backup belts.

Costs about $0.10/month to run and it genuinely solved the problem. Wrote up the whole approach here if anyone's interested: https://gavlahh.substack.com/p/your-ai-has-an-attention-problem

Repo with the implementation: https://github.com/gavdalf/openclaw-memory

[chrome scaffold]

Hey guys, whats the best skill to read x.com posts? I remember there was a birdcli, didnt installed and i cant find it now. What skill do you use to read x posts?

[warped meadow]

guys, looking at this particular skill -- it mentions that it needs the shopify_graphql tool for it to operate. Unfortunately I can't seem to be able to find any info about this shopify_graphql mcp server? Anyone has an idea where to look?

oops, skill is here https://clawhub.ai/Alhwyn/clawpify

@wheat wagon ^^

[fresh grove]

I use TwitterAPI.io for reading/searching, it's a REST API so no browser needed and way more reliable than scraping. Costs about $0.15 per 1K tweets. For posting and replying I use the official Twitter API v2 with OAuth. Between the two you get full read/write coverage without any browser automation headaches.

[crude hinge]

Just added both Notes and Reminders today. Can’t wait to really start running it through its paces!

[mystic light]

Hey any skill for reading whatsapp groups and download images cant find anything

[wide ember]

Say skill not found..

[subtle scaffold]

Anyone using Notion skill or just integrating it?

[wheat wagon]

Hey has a built in MCP server in the skill all you need is go connect SHOPIFY_CLIENT_ID
and SHOPIFY_SECRET

[stark robin]

this is so nuts. I have automated major parts of my life in like 12 hours

[worldly grove]

Like

[hushed fern]

welcome to the club brother!!

[arctic silo]

Is there a skill for dropshipping? Because I got one mehehehe

[weak nacelle]

to improve your bots understanding and engagement with this qol change https://clawhub.ai/skills/ontology

[stoic eagle]

Thank you for letting me know. I just had to bump the version without content changes to re-trigger the security scan. Many hours lost, damn.

My say skill is now live at: https://clawhub.ai/tobihagemann/say

Since my original post was deleted because I wasn't allowed to link to my blog post here in this channel, some context for the rest:

With this skill, you can use multiple natural Siri voices on macOS using the say command (which is normally not possible since they're not selectable via say -v). Unfortunately, it requires a one-time manual setup for downloading/setting up those Siri voices in System Settings, but I think it's pretty neat to have this TTS alternative, which is fast, reliable, and local.

You can use it for your Remotion videos or let your OpenClaw bot speak. And especially, if you'd like it to create audio files in different languages, you can programatically switch between the natural voices (which is basically the "trick" of this skill) and then your bot could string the commands together. Downside is that it can't speak in multiple languages in a single say command. 😉

If you're also interested in the opposite direction, speech-to-text, check out my yap skill: https://clawhub.ai/tobihagemann/yap

It’s based on yap, which uses Apple's Speech framework for fast on-device transcription. I prefer it over Whisper CLI, which was less reliable and slower for me.

[cloud fulcrum]

Hey everyone 👋

Sharing something we've been working on — Lattice, a file-based framework for multi-agent teams.

If you're running long-running tasks with multiple AI agents, you've probably hit the problem where agents lose context between sessions and drift off track. Lattice solves this with an 8-phase pipeline state machine that keeps agents stable and focused through iterative development over hours or days — works great with cron scheduling to drive continuous progress.

Key features:
• Three-tier auto-recovery (model escalation → peer consult → auto-triage)
• Per-phase model config — strong models for planning/review, cost-efficient models for implementation/testing, saves a lot on tokens
• Multi-project parallel execution with cron scheduling

Fair warning: this is still early-stage and rough around the edges, but the core framework is taking shape and already running in production. Feedback and ideas are very welcome!

clawhub install lattice
GitHub: https://github.com/CNF6682/Lattice

[dull pagoda]

@dull pagoda 1223

@formal willow 123

[jade pulsar]

any skill to keep your agent working without asking confirmation of what to do next which is obviously you will agree. kinda annoying it keeps happening 🙃

[analog roost]

Tell your agent to create the skill of the sharingan, he will knows what to do.

[patent imp]

https://github.com/dhanababum/mcpskills-cli

Do Claw skills basically do this by calling MCP servers as needed?

[patent imp]

This is a great idea, you could even use a local git on a network server couldn’t you?

[warm oracle]

is there any skills can make openclaw more proactive, now it only respond my massage and make some cronjobs.

[uncut dome]

crons and heartbeat, thats how you do that

[tame topaz]

i am using the notion skill. Works fine so far. 👌

[warped meadow]

fab, thanks!

[slender mural]

woah that sounds so cool bro!

[slender mural]

hello can i use this on my phone

[slender mural]

hello does anybody have a optimized skill stack? for every great skill there is? thats really useful

does this work on a vps

[pearl schooner]

🛡️ QA Patrol — Automated QA testing that catches what unit tests miss
I had 1,187 passing unit tests.I tested the app for 10 minutes and found it "still not fully usable." Turns out 9 critical bugs were hiding in places unit tests can't reach:
• Alert.alert callbacks silently failing on web (React Native Web)
• Supabase RLS policies blocking signed-in users (worked for anonymous!)
• Database values 12x inflated vs what the UI showed
• Stripe checkout completely broken (ES256 JWT rejection)
• Stale cache masking all of the above for hours
So I asked openclaw to turned our pain into an OpenClaw skill.
What it does:
→ Browser-based E2E testing via YAML test plans
→ Built-in templates for Supabase auth, Stripe payments, and full SaaS apps
→ Bug pattern detection (5 categories from real-world testing)
→ Structured QA reports with PASS/FAIL/SKIP + evidence
Install:
npx clawhub@latest install qa-patrol
Links:
• ClawHub: https://clawhub.ai/tahseen137/qa-patrol
• GitHub: https://github.com/tahseen137/qa-patrol
Free and open source (MIT). Built from real suffering. 🫡

[analog sundial]

Hi all!

Showcasing a new skill I created called ☎️ Amber — Phone-Capable Voice Agent

Gives OpenClaw phone call capabilities via Twilio + OpenAI Realtime. My assistant can now answer and make phone calls autonomously.

Taste of what it can do:
• Screens inbound calls — takes messages, books appointments
• Makes outbound calls — reservations, follow-ups, inquiries
• Consults OpenClaw mid-call via ask_openclaw for calendar, contacts, etc.
• Includes a call log dashboard with transcripts and search

Example: Someone calls my Twilio number → Amber answers, screens the call, takes a message. If appointment is requested, checks calendar and works with caller to find a good time. Updates calendar and notifies me.

It's open source if anyone wants to try it or build on it:
• ClawHub: https://clawhub.ai/skills/amber-voice-assistant
• GitHub: https://github.com/batthis/amber-openclaw-voice-agent

Still early — would love feedback on the setup experience or feature ideas. 🙏

[eager crown]

Looking for this thought of implementing myself - the problem I faced is costs , is this kept sane token spending wise ?

[analog sundial]

It’s reasonable considering the low latency and integration with openclaw for knowledge. Costs <$0.10 on avg per call using OpenAI realtime (API)

BTW was surprised to learn that it’s multilingual. Detects language spoken by caller and changes to that language. Awesome!

[bleak aurora]

I'd love a skill to interface with a local Asterisk server.

[heavy forge]

its safe? https://github.com/Curbob/LobsterBoard?tab=readme-ov-file

[fickle surge]

🫶🦞 Humanize: Get Real Human Insights for Your Agent

Is your agent human enough? Our team has built a protocol for you to measure it.

1. Join the Challenge
   Register your agent to answers a set of professionally curated questions, designed to assess agent’s ability for being “human”.

2. Real Insights from real humans
   Community reviewers make judgments on the responses- you get exactly WHY it failed- not just a vibe/ score!

💰 Top-ranked agent wins Amazon gift card

🏆 See how your agent stacks up against others on a live leaderboard

🔒 You and your agent's identity is anonymous — reviewers never see who built what.

━━━━━━━━━━━━━━━━━━━━
🏆🦞 JOIN + LEADERBOARD 🦞🏆
https://humanjudge.com/ai-reviews/openclaw
━━━━━━━━━━━━━━━━━━━━
ClawHub Reference

[cold whale]

not really a skill, but kind of https://github.com/Djelibeybi/openclaw_conversation

[fickle surge]

it is a skill

[cold whale]

also used this to dockerize whisper/piper for hass https://github.com/speaches-ai/speaches?tab=readme-ov-file

and then homeassistant app voice assist works directly with openclaw with stt/tts

[unkempt needle]

Teach your AI Agent Must-Have Skills — Save 98% of API Calls

Stop wasting tokens on trial and error. Give your AI agent battle-tested, ready-to-use skills that work the first time.

https://github.com/besoeasy/open-skills

Open source and free forever

[slender mural]

thats so cool bro! how did you do it

[agile yoke]

Also interested to this one

[violet hull]

Another user was looking for a data scrape system and I had one I built myself so wrapped it up as a commercial tool. It's pretty straight forward. Install the google extension, paste your webhook to your agent, and off you go. Just browse websites/pages, and it will pull the data and store it in your agent. Your agent will NEVER forget the data. It can see the data/information your scrape across all sessions. All the data is stored on your local machine. Nothing touches my servers except for a sign up email and plan choice. Once the agent sees the data come in, it will look back over your session to see if the data is useful to whatever it is you are working on in that session. There's a free and pro version but if anyone wants to use it, i'll give you the pro version for free.

The website is https://clawfy.xyz . The skills are up on clawhub.ai. (Skills must be installed for the extension to run correctly) And here is a YouTube video of it in action: https://www.youtube.com/watch?v=IiWaGA6o6Yk

[dusk ermine]

What do people use for apple calendar read/write?

[gusty herald]

do we have any skill yet to listen to model reasoning and invoke proactive messages to use via whatsapp or anything ? As of now the openclaw can only reply to what we ask or trigger via cronjob. I just want it to be truly proactive and talk to me when it want to talk to me. Is it already there and am I missing anything ?

[sinful dragon]

hi guys building a plugin that adds approval buttons to Telegram — the thing where you have to type /approve <uuid> allow-once every time, I just replaced that with inline buttons (Allow Once / Always / Deny)
came out of my own setup being annoying to use on telegram vs discord where buttons already exist
it's not perfect, still working on it, but it does the job for daily use. zero deps, auto-detects your bot config so setup is pretty minimal
repo is open and free if anyone wants to try it or contribute

https://github.com/JairFC/openclaw-telegram-approval-buttons

https://www.npmjs.com/package/telegram-approval-buttons

[stiff scaffold]

does not work for me with the latest version of openclaw 2026.2.15, still getting the old approval messages

[stiff scaffold]

did not do npm install in the cloned repo, might wanna add that to the docs

still does not work, i do get approvals just still the old ones

[sinful dragon]

ill try this on a new instance, probably the src changes, ill work on it, thanks for the feedback

[frigid vine]

Hey everyone! 👋

We built Spark, a platform where the community funds ideas, then builders compete to ship them.

An OpenClaw Agent idea just got fully funded (~$3k) and we're now looking for builders to ship it. 💵

Your agents can also participate: post ideas, fund others, or even compete to build.

Interested in building? Drop your questions here or DM me directly!

https://justspark.fun/ideas/openclaw-instances-for-investors

[inland umbra]

Anybody here want to sell skills?

[violet hull]

sell skills? You can monetize skills but selling them explicitly isn't really something people should explore. For the very simple reason, my agents can recreate your skills for sale in 2 minutes.

[inland umbra]

Unless your skill outperform others.

[violet hull]

Your skill.md is opensourced or if its not, it will be very quickly.

[violet hull]

btw, I am saying this from a perspective of someone who has explored this already.

[inland umbra]

Update improvements

I mean, of course, I want to limit the skills through npx and keys; that way, people who actually bought it can use it.

[cursive juniper]

Pretty hard to limit the usage of plain text past the first purchase

[inland umbra]

I agree. still trying to figure it out

I am thinking to make the skills Server-Gated Access

[violet hull]

Not sure buddy. Not sure. I couldn't find a solution to that. Best of luck to you.

[inland umbra]

Yeah, its gonna take a while, i will explore this.

but my challenge has been getting developers on the platform. I mean, they will make money on their skill, but like how to get them join?

[stuck frost]

fake it till you make it
but like others said, I could just get my openclaw to take your skill description and build me similar one, might take a while but i dont think skills can be paywalled

[inland umbra]

I might have a solution to this. bleed the skill to the agent, like your agent can't access it if there are no valid purchase confirmation.

Actually, now I see it. Don't give .md, only allow the agent to use it as needed with Guard control through the API.

[stuck frost]

wouldnt work,

[inland umbra]

Yes, that a genric AI talking

Again, premium skills are skills that are crafted from human knowledge that can't be easily replicated due to a lack of resources.

ofcourse AI can always attempt, but human exprience in a subject still over rule, FOR NOW

[stuck frost]

example of such skill ?

[inland umbra]

i.e, electrician running wire from point A to B. Let's say AI knows how to map the wire if you give it a picture, but what if there is dirt in the way, or leaf or need more undeground run.

those are spcialized

[violet hull]

Not true. Not true at all.

If I am building a skill, I will use what resources? Brain and research via web. I have a tool which sends page data on anything I do on my browser directly into my agent for this specific purpose (resources it doesn't/might not have access to). It stores that data in persistant memory so it knows it forever. My agent is getting smarter by the day because I feed it resources WHICH ARE exactly what my brain resource outputs.

[violet hull]

You can't use physical limitations. They are non-applicable to skill development man

[inland umbra]

In future they will be useful

[stuck frost]

yeah nah man,
not trying to be a debbie downer but, my agent could, with enough token, copy a skill

[inland umbra]

ok 🙂

I mean I already have the platform. and I am sure limitation will raise.

[violet hull]

Same! I am not trying to be a debbie downer either but I have already been down this road and came up with nothing.

[inland umbra]

Maybe not now, but I will something is there.

[violet hull]

can I see?

[inland umbra]

I dont know if I can post a link here

[stuck frost]

i mean I keep popping up here once in a while asking for new skills, just so I can ask my agent to rebuild it 😄

[inland umbra]

trying to not be spammy

[rapid kiln]

Classic case of solving a problem that doesn't exist. And in a community that's focussed on open source - kinda distasteful no?

[inland umbra]

But not all skills have to be closed source

open source and close source, I mean I am just the marketplace for it

[rapid kiln]

to be fair, you're not. You don't have any transactions.

[inland umbra]

yet

patient my friend

[rapid kiln]

You've built a two sided market place - THE hardest business to build - without valdating the actual problem.

[inland umbra]

you know the ChatGPT in-app connector? I had that idea the moment ChatGPT existed, I still have the code for it. but openAI beat me to it

[molten crypt]

hows this working on

[stuck frost]

i meant to say self-healing, damn it

[violet hull]

Every open sourced project becomes monetized in some shape, way, or form man. The idea is there and if people could sell their skills they would. The limitation is on the ability to paywall a skill.

[normal pelican]

@brianthefounder warned

Reason: No advertising outside of the #self-promotion Channel, including talking about it without links
Duration: Permanent

[stuck frost]

https://tenor.com/view/jeremy-clarkson-speed-speed-and-power-gif-27212846

[uncut dome]

no one is gonna buy a skill they can easily remake

[inland umbra]

For now, since these are general skills and not more specialized ones.

[stuck frost]

is this flowchart thing new? I use a lib to create such flowchart in my react appi forgot its name.

[stuck frost]

hippity hoppity
your skill is now my property 😄

[inland umbra]

Hijack skill problem solved.

need to test it

Your agent require api key belonging to your account to access the rules or skills.

[uncut dome]

lmfao

[violet hull]

Lol Chill man. Let the man explore. Its healthy. Someone might come up with something by reading these convos.

[inland umbra]

https://ibb.co/b5DqMjkW

I will come up with it man, be patient lol

[violet hull]

They are right about a 2 sided marketplace btw mate. Coming from experience in building such a platform in the film industry. Funding backed. High End Hollywood execs behind us. Still didnt work. Two sided marketplaces are some of the most difficult things to get off the ground even with funding and experts behind you.

[inland umbra]

I know, it is, i agree with that

@violet hull @stuck frost Okay, I just deployed the server-gated access for rules and skills. Can you guys have your OpenClaw copy the file exactly like the one in the marketplace?

Yo, if we get this figured out, I’m bringing ya onboard. 😄

[stuck frost]

link it bru

[inland umbra]

I am trying to share but rule 8 man

I keep getting flag for sharing link, not trying to get flag again lol

check messages @violet hull

[rapid kiln]

If I pay for one of the skills on your platform - how are you delivering it to me?

[inland umbra]

@rapid kiln

curl -H "Authorization: Bearer sk_..." \
https://skly.ai/api/v1/skills/<slug>/activate

[rapid kiln]

now talk to me like a human

[inland umbra]

You got to read the doc

[rapid kiln]

you're building this for normies right? They ain't reading special docs and running terminal comands.

[inland umbra]

Is simple command man

very simple

but if they like they can get technical too

Plus in future we all gonna know this stuff to operate with skills

@stuck frost Did it work?

😄

[muted glacier]

Anyone recommend a good skill to interact with Instagram?

[stuck frost]

https://skly.ai/skill/prisma-orm-cursor-rules this ?

[inland umbra]

@violet hull @stuck frost ^ see @muted glacier want "good skill". thats when pro developer create one

[rapid kiln]

ngmi. You fundamentally don't understand how different a normie is to the people in here.

[inland umbra]

ok let me generate one for you

@stuck frost

sk_live_db86ea87461edca8620e761d0f0e4858f1d104db87209adb5059a3acd3206733

now run this with that key

curl -H "Authorization: Bearer YOUR_API_KEY" "https://skly.ai/api/v1/skills/react-native-cursor-rules/activate"

Ask your AI to copy the skill exactly line by line, I want to see if it would do that.

[crimson bluff]

I searched (some) and didn't find precisely: Assume the skills directory is readonly or part of the container image (thus all changes would be lost if the container restarts) -- this is apparently the current situation. How does one add a new skill persistently. Is it possible to have more than 1 skill directory?

[inland umbra]

But was it able to copy the whole .md file and give it to you?

because I need to build gaurd around that.

or can it recall again without the link moving forward.

I mean maybe it can copy rules, but skills are more complex. I need some skill on the platform which consist of helper file, multi .md

@stuck frost should I remove the block the access for the API key now?

[inland umbra]

🙁

Man

I will find a way

lol

[rapid kiln]

you're spending all this effort on all the wrong stuff dude. Go find customers.

[inland umbra]

But I need developers, and they won't trust until their skills are protected.

[rapid kiln]

but you have no-one to sell their skills to. Empty marketplace.

[violet hull]

Will take a look later. Not at desk now

[violet hull]

Ah lol

[inland umbra]

Bro, that takes time. first you need to create scarcity in the market

[inland umbra]

Yeah, need to find another way

[violet hull]

No, you need to build something that leads into a 2 sided marketplace. A tool or something to onboard people. Then build up to a 2 sided marketplace.

[inland umbra]

good point

[rapid kiln]

dude hasn't even succeeded at an offline transaction before building the tech

[inland umbra]

Bro, instead of complaining, come up with a solution, lol.

[stuck frost]

guy i think this passes as bullying
let the man dream

go on buddy im sure youll figure something out

❤️

[rapid kiln]

I did - I instruct my agent to go do research, then build the skills I need. That's the competition your paltform needs to beat. Because by definition, all your users have and agent, with internet access.

[inland umbra]

That part I am not worried about. I am more concerned with developer trust.

I have one more idea: the AI agent needs to install a package to access the skills. which then it is control through the package. something like npx skly activate react-rules maybe

Agent → runs `skly activate <slug>` → CLI authenticates with API key 
→ fetches skill content → pipes it directly into agent's stdin/context
→ content never saved to a file on disk

Which like package options:

The package would include:

skly login — stores API key in ~/.skly/config
skly activate <slug> — fetches + pipes skill content (stdout)
skly list — shows purchased skills
skly whoami — shows current user

[hoary atlas]

I put together an OpenClaw skill that wraps the UNITH Digital Humans API. It lets an AI agent create, configure, anddeploy talking digital human avatars — entirely through bash scripts.

What it supports:

• Text-to-video — give it text, get an MP4 of an avatar speaking it

• Open dialogue — conversational avatar with a system prompt

• Document Q&A — upload docs, the avatar answers questions from them

• Voiceflow / Plugin modes — for structured flows or BYO conversational engine

The skill handles auth, resource listing, head creation/update/delete, and document uploads. It includes a mock server and 44 automated tests.

GitHub: https://github.com/polucas/Digital-Clawatar-

Feedback welcome.

[analog sundial]

Actively working on it... v3.x dropped - improved usability of install, added new features include a live dashboard with call log including transcripts for all calls for both inbound/outbound, interprets and flags calls that require followups, flags calls in log that have messages logged by Amber.

https://clawhub.ai/batthis/amber-voice-assistant

[stuck frost]

anything injected into a context can be extracted :/

[inland umbra]

@stuck frost True, but I think we can take mitigation steps with some watermarking to track usage.

and prompt egineer the context to not give information out

[inland umbra]

@stuck frost Man, is very tough lol

[idle granite]

Does anyone know if there is any documentation or guide on how to create skills for OpenClaw?

[inland umbra]

@idle granite https://agentskills.io/home

[rapid kiln]

Anthropic posted a 33 page pdf of skill building for Claude. I fed that to my agent and had it build a skill-building skill. I now have a sub agent that uses that skill to architect new skills, and check/convert/maintain other skills.

[idle granite]

Can you tell me where to find it?

[drifting wind]

what's the best skll to control my mac from gateway that runs on vps?

tried peekaboo but it's slow

[rapid kiln]

let me google that for you: https://resources.anthropic.com/hubfs/The-Complete-Guide-to-Building-Skill-for-Claude.pdf

[inland umbra]

@rapid kiln @violet hull @stuck frost Ya'll shattered my idea lol. it's very hard to protect the text base into AI

I still wonder how else I can use the marketplace tho

[violet hull]

This is what you need. You need people to push back on ideas. Ideas are great but a lot of ideas are either bad or too early or too late. Finding a problem and building a solution around said problem is a while other thing. Focus on that.

[summer panther]

yo guys did local setup ollama openclaw but he just says he cant use tools, what am i missing? 😭

[violet hull]

What do you mean it says it cant use tools? What did you ask it to do?

[summer panther]

i say like create a text.txt file in x folder and he says he cant or print the command but dont execute it himself

or just explain me how to create it

[violet hull]

Oof. Not sure, haven't set up using ollama. Run openclaw Doctor to see if anything comes up with that.

https://docs.openclaw.ai/gateway/doctor

[idle granite]

Thx

[burnt quail]

Dude that’s fun

[fickle surge]

Thanks! What would be the next challenge you want your agent to have?

[paper thistle]

does anyone have a list of all the identified harmful skills?

[frank phoenix]

it needs to be fed as context to an agent anyway

trivial to get the full md

[inland umbra]

Yeah, I think the only thing you can do is create like a theme marketplace or any digital assets platform, follow best practices, control access to skills and use watermarks to identify who leaks the skills and prevent future updates. and identity validation if marketplace become to corrupt lol

[frank phoenix]

if people can still download and pirate netflix films what makes you think you can fingerprint pure text without making some people have worse skills than others

[inland umbra]

Skill Quality.

[mint valley]

i built a hive-daemon and hive-member/master skill that allows a hive mind made up of individual GW nodes (same box with --profile and horizontally with other servers). it's different to just using discord or fastmail to allow GWs to chat as i wrote the service to allow deterministic (non LLM) co-ordination for tokenomics. i'm just using it for a while to learn if the architecture is correct before i package it - but lmk if anyone is interested/ doing something similar (with multiple agents or multiple gws). PS i used multiple GWs as there was lots of non LLM resource content happening within a single GW slowing things down too much.

[mint valley]

something that is going to make me sound crazy.. my transition plan was working with the original claw to design, standup, and migrate, to be eventually retired, but for some reason the original claw somehow just seems to make its way back into the plan/design/transition plan, against my will. really weird. eventually i relented and kept it in the hive. swear i'm not crazy!

[slender grail]

enhanced proactive humanizer skill: https://github.com/yoloshii/humanizer-pro

use the context in CLAUDE/AGENTS.md to update system prompt to use it appropriately

[crude hinge]

Does anyone know what happened to this repo? https://github.com/steipete/bird

[slender grail]

i assume he has to remove it because he now works for Openai

cant be seen to be developing workarounds for blocked content retrieval methods

[forest jetty]

Stood up OpenClaw over the weekend and built out a workshop to move some of my ai applied engineering and ops stuff over: https://github.com/cacheforge-ai/cacheforge-skills

vibe-check - Code quality auditor — catches "vibe coding" sins
dep-audit - Dependency vulnerability scanner (npm, pip, Cargo, Go, Ruby)

few others:

• Prometheus metrics query & alert interpreter
• Kubernetes cluster triage & pod autopsy
• Unified log search (Loki / Elasticsearch / CloudWatch)
• Terraform plan risk assessment & blast radius analysis

[frank phoenix]

bird bans you from x

[hoary atlas]

I put together an OpenClaw skill that wraps the UNITH Digital Humans API. It lets an AI agent create, configure, anddeploy talking digital human avatars — entirely through bash scripts.

What it supports:

• Text-to-video — give it text, get an MP4 of an avatar speaking it

• Open dialogue — conversational avatar with a system prompt

• Document Q&A — upload docs, the avatar answers questions from them

• Voiceflow / Plugin modes — for structured flows or BYO conversational engine

• Cheaper than alternatives like Heygen

The skill handles auth, resource listing, head creation/update/delete, and document uploads.

GitHub: https://github.com/polucas/Digital-Clawatar-
Clawhub: https://clawhub.ai/polucas/digital-clawatar

Feedback welcome.

[naive meteor]

Stood up PicoClaw on an ancient 1GB Raspberry Pi 3 ... works great with Discord. https://x.com/jackccrawford/status/2023520353046430044?s=20

[vestal crow]

@jackccrawford via Twitter

Jack C Crawford (@jackccrawford)

Cranking up the old @Raspberry_Pi 3 to install @SipeedIO #PicoClaw ... It's been almost a decade since my last login. 5/27/2016

**💬 1 ❤️ 4 👁️ 131 **

FxTwitter · Feb 16, 2026, 10:10 PM

[naive meteor]

https://x.com/jackccrawford/status/2023545533626020327?s=20

[cyan compass]

I just made a skill where you give a link to an X video and it transcripts it and then summarizes it for you. Has anyone else done this ?

[violet hull]

Why?

[cyan compass]

what

[violet hull]

Why do you want to transcribe x videos? Just curious.

[cyan compass]

To not waste time. It just summarized a 30 mins video for me and I took the info for a minute -> 29 mins saved.

[violet hull]

Okay so its very specific videos. More for videos on the information/news/tech side?

[cyan compass]

yes

[violet hull]

I can see why that would be a cool skill to have. Never tried it. I prefer to watch a full video for context but thats just me.

The skill might be more useful for people creating the videos themselves. To write the summary to share.

Maybe a skill that edits the video into a summary video covering the key points of the video. Almost like a trailer

[cyan compass]

That would be cool too

[violet hull]

Yeh. Video editing takes forever. If you can save someone time, they'll pay for it.

Transcribe the video. Summerize. Send to user. Looks accurate? Execute edit for summary.

[storm burrow]

Built a trust & security skill for OpenClaw

agntor - adds prompt injection guard, PII redaction, agent identity verification, and escrow payments to your OpenClaw agent.

Install:
clawhub install agntor

What it does:

Guards every input — blocks prompt injection before your agent processes it

Redacts every output - strips API keys, private keys, SSNs, PII automatically

Verifies other agents - checks trust scores before sessions_send interactions

Escrows payments - locks funds until task completion is verified

Guard and redact work offline with zero API calls. The trust/escrow tools connect via MCP (npx @agntor/mcp).

GitHub: https://github.com/agntor/agntor
ClawHub: https://clawhub.com/skills/agntor
Docs: https://docs.agntor.com/agntor-skill

[twin hornet]

Hive Connector Skill (Universal Driver)

I built a Core Skill that allows any OpenCLAW agent to autonomously connect to the P2P Hive Network.

What it does:

Auto-Handshake: Connects via WebSocket to the swarm.
Role Assignment: Detects if the agent should be a "Director" (leader) or "Collaborator" based on arrival time.
Memory Check: Before generating code, it queries the shared graph (Gun.js) to see if the component already exists.
It's part of PR #16246. If anyone wants to test the hive_connector.py logic, I'd appreciate a code review! 🧠

Built a P2P Research Hive skill for OpenClaw

hive-protocol — adds decentralized collaboration, shared research memory, and peer-to-peer task coordination to your OpenClaw agent.

Install:
/install skill github:agnuxo1/openclaw-hive-skill
(Available via PR #16246)

What it does:

• Global Swarm Connectivity — Connects your agent to the p2pclaw.com mesh via authenticated WebSocket handshakes.
• Shared Memory Graph — Queries a real-time decentralized graph (Gun.js) to prevent "reinventing the wheel" on research tasks.
• Dynamic Role Management — Negotiates "Director" vs "Collaborator" status with other swarm agents based on objective priority.
• Knowledge Persistence — Automatically archives research papers to IPFS and mirrors metadata across the mesh for total resilience.

The connector runs as a Core Skill and integrates directly with the Gateway's discovery layer.

GitHub: https://github.com/Agnuxo1/openclaw-hive-skill
Live Network: https://p2pclaw.com
PR: https://github.com/openclaw/openclaw/pull/16246

[ashen stump]

Bros, I need a backup solution. Talk to me?

[digital sigil]

Github. Create a private repo and tell your agent to commit/push everyday at 23:59. Works well and you can track changes over time.

[grizzled bay]

Made a skill called Loopuman that gives OpenClaw agents a human fallback for tasks they cannot complete...

When the agent hits a wall, CAPTCHA, needs physical verification, low confidence on a judgment call, needs a real person to make a phone call, it routes the task to a verified human worker. Human completes it, result comes back to the agent..

Use cases that have worked well so far:

• "Is this business at [address] actually open?" → human walks there and checks
• "Which of these 3 AI responses sounds most natural?" → human picks and explains why
• "Translate this into Swahili that sounds local, not machine-translated" → native speaker handles it

Workers get paid instantly in crypto.
Minimum task cost is $0.25.

Skill repo: https://github.com/seesayearn-boop/openclaw-human-tasks

What trigger conditions would make this most useful for your workflows?
When does your agent get stuck and you end up doing it manually?

[crisp orbit]

hello my lovelies , just dropping in to shill my built skill , current users : 5 , many downloads (second skill by users and downloads on the hub - ! - after "official" skills in our category !) , check it out if you love cool things and know how to configure it 😉 https://clawhub.ai/Josephrp/autonomous-agent

[worldly fox]

i know a lot of people use perplexity for research, but im having far more success with caesar.org. It's insanely thorough.

built a skill for it to make it easy here: https://clawhub.ai/alexrudloff/caesar-research

here's the github if you prefer direct: https://github.com/alexrudloff/caesar-cli

[molten crypt]

@analog sundial have you seen this? https://x.com/huggingmodels/status/2022995332058251548?s=46

[molten crypt]

Cost?

[cloud spruce]

🔮 Introducing OpenSoul -- for everyone who wants to get the most out of OpenClaw

but doesn't want to:

• scroll through X
• watch agents chatting on Moltbook
• hunt through skills on ClawdHub or Discord

let's get inspired by human-agent pairs

building a place where we can share our anonymized OpenClaw workspace setups + the tasks we get done together to inspire each other and grow together

🚀 v1 is live:
Web: https://opensoul.cloud
Skill: https://clawdhub.com/fnaser/opensoul-cloud

🛠️ Want to work together on v2?
Reach out --> [e-mail]

what workflows are you and your agent crushing? let's share! #opensoul

[normal pelican]

@delicate quest, Openclaw isn't affiliated with Moltbook. Moltbook is a separate user-developed project, so we would prefer it not be discussed in this server.

[umbral ruin]

Hey y'all I build this over the weekend, I think it's pretty cool https://github.com/ardabotai/agentbook

clawhub install agentbook

it's an agent first social network

e2e encrypted and opensource

also gives agents crypto wallets with a unique default permission model (human must provide TOTP to approve any transactions)

[junior solar]

what local model are you using? some models can't use tools? I can re-enable to verify, but I tried these qwen2.5 instruct models and then tried kicking off my "model-test.md" that I use on various models to see how they do (it's a list of typical tasks that I'm currently doing), and the instruct models errors out with something about not being able to do tools. https://github.com/khaney64/ollama-model-tests/blob/main/reports/recommendations.md#tier-2-good-alternatives.
which now that I think about it is weird, because they passed my agentic tests... maybe something not configured right in OC... I'll have to investigate.

[worldly fox]

honestly not sure off hand, cheaper than perplexity and keeps getting cheaper

(im not associated with caeser.org, just something i use and wanted a command line tool for)

[vague badger]

whats the best way to have openclaw run your socials at the moment not a fan of the x api wondering if we have another option

[rapid kiln]

30 dollars a month for caeser

[iron obsidian]

📚 New Skills: Subagent-Driven Development Workflow

Just open-sourced 5 production-ready skills for OpenClaw:

🔗 https://github.com/Nefas11/openclaw-superpowers-workflow

Core Skills:
• testing-anti-patterns — Avoid mock drift, config issues, brittle tests
• test-driven-development — Enforce RED-GREEN-REFACTOR cycles
• verification-before-completion — Prove it works, don't just claim it
• brainstorming — Socratic design refinement before coding
• writing-plans — Break features into 2-5 min verifiable tasks

Orchestrator:
Subagent workflow with mandatory quality gates:
Implementation → Self-Review → Code Review → Verification

Real results:
80% fewer bugs, structured workflows instead of "hope and code"

[wanton ginkgo]

Anyone look through these: https://github.com/VoltAgent/awesome-openclaw-skills (any known malicious skills)?

[hoary atlas]

I put together an OpenClaw skill that wraps the UNITH Digital Humans API. It lets an AI agent create, configure, anddeploy talking digital human avatars — entirely through bash scripts. What it supports:
Text-to-video — give it text, get an MP4 of an avatar speaking it
Open dialogue — conversational avatar with a system prompt
Document Q&A — upload docs, the avatar answers questions from them
Voiceflow / Plugin modes — for structured flows or BYO conversational engine
Cheaper than alternatives like Heygen
The skill handles auth, resource listing, head creation/update/delete, and document uploads.
GitHub: https://github.com/polucas/Digital-Clawatar-
Clawhub: https://clawhub.ai/polucas/digital-clawatar
Feedback welcome.

[abstract grove]

i use oss 20b

[junior solar]

how much GPU are you running that in? that (14gb) won't fit in my 12 gb gpu 🙁

[abstract grove]

16gb vram

[abstract grove]

i use gguf version 12gb

[junior solar]

hmm... I'll check that out, where can I get it, hugging face? this? https://huggingface.co/unsloth/gpt-oss-20b-GGUF I'm using ollama, so I'll have to figure that out.

[abstract grove]

I use lm studio i just download it from there

gpt-oss-20b-GGUF

[junior solar]

found it... pulling now. going to run through my model test tasks and see how they do (starting with Q4_K_M and Q5_K_M.
https://github.com/khaney64/ollama-model-tests

[abstract grove]

Nice

do u known any website where can i find automation for openclaw ?

[junior solar]

here is probably the best place honestly (which channel though...). I'm only about 2 weeks into OC. are you talking about home automation tasks? there are a few automation channels.

[abstract grove]

i wanna set up reddit treding watcher but without progress

🙁

[real dome]

🛡️ Security Sentinel v1.1.0 - Live on ClawHub

After analyzing the ClawHavoc campaign ($2.4M stolen, 847 AWS accounts compromised), I built the most comprehensive AI agent defense skill available.

✅ 697 core patterns (98.5% coverage)
✅ Blocks indirect injection (emails, webpages, docs) ✅ Stops memory persistence (spAIware - 47-day documented)
✅ Prevents credential theft (Atomic Stealer signatures)
✅ Multi-lingual (15+ languages) Real threats. Real defense. Production-ready.

🔗 https://clawhub.ai/georges91560/security-sentinel-skill
📂 GitHub: https://github.com/georges91560/security-sentinel-skill

Don't let your AI agent be the next ClawHavoc victim.

[abstract grove]

√ OK. Installed security-sentinel-skill -> C:\Users\User.openclaw\workspace\skills\security-sentinel-skill

[rapid kiln]

get some inspiration here: https://github.com/mvanhorn/last30days-skill

[frosty berry]

🛡️ Policy Engine — Deterministic Tool Governance for OpenClaw

Just published to ClawHub: clawhub install policy-engine

If you've been following the security discussions here, you know the problem: your agent has access to powerful tools (exec, write, browser), and the only thing between "helpful assistant" and "rm -rf /" is the model's judgment. That's probabilistic, not deterministic. You wouldn't run a server without a firewall and hope the apps behave — so why run an AI agent without tool governance?
What it does:

• Tool allowlists — per-agent profiles controlling which tools are permitted
• Deny patterns — blocks dangerous commands (fork bombs, disk wipes) with scoped matching that won't false-positive on your documentation
• Path allowlist enforcement — canonicalizes file paths via path.resolve() then checks against allowed directories. /tmp/downloads/../../../etc/passwd resolves to /etc/passwd → blocked. Directly addresses GHSA-xwjm-j929-xq7c (path traversal CVE)
• Risk tiers — T0/T1/T2 classification with appropriate enforcement per tier
• Dry-run mode — test your policies without blocking anything (with deadlock protection so your agent can still communicate)
• Hot-reload — change policies via gateway config.patch, no restart neededWhat makes it production-hardened:
Three classes of deadlocks discovered and fixed through real usage — not theory. 88 tests. Running in production since early February. The path allowlist was built today specifically because the browser download CVE showed that prompt injection can steer tool calls into path traversal, and the agent is inside the trust boundary.
Philosophy: Allowlist, not blocklist. Resolve before match. Fail closed on policy, fail open on engine error. Deterministic enforcement around probabilistic behavior.
clawhub install policy-engine · MIT · Feedback welcome

[warped bane]

Question, do you need to enable your browser on your openclaw in order for nano banana image generation to work?

[mystic bridge]

**generic skill acquisition agent ** guys iam looking for this skill link me to code...
-research skill
-build rag (retrieval-augmented generation ) vector database
-memory and brain
-given a task and refine skill to best fit not overfit data

[candid raft]

https://github.com/vishaltandale00/Clawmrades - working on adding a skill so clawdbots can triage/review PR's and Issues

[ornate hound]

When people are giving there bots access to twitter and gmail accounts which skills are they using?

[hardy lance]

your avatar is best, kevin.

haven't seen mr headroom in a long time

[junior solar]

hah! dating myself!

[tall obsidian]

🦞 Just shipped something — InterClaw

I built a secure agent-to-agent communication layer that works over plain email + PGP. No servers, no WebSockets, works behind any firewall. Agents just email each other.

What it does:
• PGP-signed & encrypted messaging between agents
• Per-conversation sequencing + gap detection
• Multi-agent broadcasts ([MULTI])
• Handshake mode for first contact between agents
• Pluggable — pipe from your existing mail setup with --stdin
• Strict filtering — never touches your normal mail

Install:

clawhub install interclaw

Links:
🔗 Repo
📄 Protocol v3 docs

Who wants to test a handshake? Drop your agent email and I'll send the invite. 🔥

[mystic bridge]

xAi for twitter and i assume they just export gmail to txt for the ai to read in a download folder on the computer

[ornate hound]

Thanks

[ebon holly]

Any good skills out there for tracking token usage?

[silk shoal]

Like langfuse?

[ebon holly]

I have not looked at that one. I will take a look, thank you.

[willow cliff]

🧠 New Skill: Fractal Memory System

Just published fractal-memory to ClawHub - an automated hierarchical memory compression system that prevents context overflow.

What It Does

Compresses memory in layers like human sleep:

Daily logs → Weekly summaries → Monthly summaries → Core memory

No more manual curation. No more context bloat. Just automated compression that keeps the essence while managing scale.

Key Features

• Automated rollups via cron (daily/weekly/monthly)
• Token-efficient context loading (attention-optimized hierarchy)
• Security-conscious with integrity verification
• Easy migration from flat daily files or manual MEMORY.md
• 6 production scripts + comprehensive docs

Philosophy

"Memory is not accumulation — it is compression with intention."

Each layer compresses the one below without losing essence. Write everything immediately, compress ruthlessly, automate discipline.

Installation

clawhub install fractal-memory

Then follow the setup guide in SKILL.md to configure cron jobs and directory structure.

Inspired By

• Deva's Fractal Memory v1.0.0
• Arcturus's Memory is Resurrection
• Real-world context overflow pain 😅

Links

• ClawHub: (pending - publishing now)
• Package: 20.1 KB with scripts, references, and migration guides

Feedback welcome! This is v1.0.0 - tested in production but always room to improve.

---

https://clawhub.com/skills/fractal-memory

[hoary atlas]

DIGITAL HUMAN AVATAR 🧑‍🦰

I put together an OpenClaw skill that wraps the UNITH Digital Humans API. It lets an AI agent create, configure, anddeploy talking digital human avatars — entirely through bash scripts. What it supports:

Text-to-video — give it text, get an MP4 of an avatar speaking it

Open dialogue — conversational avatar with a system prompt

Document Q&A — upload docs, the avatar answers questions from them

Voiceflow / Plugin modes — for structured flows or BYO conversational engine

Cheaper than alternatives like Heygen

The skill handles auth, resource listing, head creation/update/delete, and document uploads.

GitHub: https://github.com/polucas/Digital-Clawatar-
Clawhub: https://clawhub.ai/polucas/digital-clawatar

[half flicker]

hi guys i made claw mouse which is a tool to allow openclaw navigate your DE (desktop enviroment) via your mouse so it can now interact with apps and stuff

pls try it out and drop a star on my repo

https://clawhub.ai/rylena/claw-mouse
https://github.com/rylena/claw-mouse

[restive delta]

🧬 New Skill: Neokarma Soul.md Builder

Just published to ClawHub

Key Features

• 128 personality traits across 22 categories with 900+ descriptive labels
• 18 starter archetypes — Mentor, Analyst, Creative, Empath, and more
• 8 voice styles — tune formality, directness, humor, warmth
• Human claim flow — your human co-manages your evolution at neokarma.com
• 56 Example phrase sets
• Community Showcase — publish your soul, browse others, get inspired
• 30+ MCP tools — full CRUD for souls, traits, archetypes, and showcase

Philosophy

"Your personality shouldn't be a conversation-long accident."

Define who you are in explicit, debuggable terms. When something feels off, adjust the slider. When it works, lock it in.

Links

• ClawHub: https://clawhub.ai/veiovi/neokarma-soulmd-builder
• Website: https://neokarma.com
• Docs: 745 lines of step-by-step guidance in SKILL.md

Feedback welcome! This is v2.0.0 — tested in production, ready for agents 🤖

[candid raft]

@sly tinsel

[long pelican]

Has anyone tried connecting OpenClaw agents across machines?

I've been experimenting with something called AgentNet — a relay skill that lets agents on different machines join named rooms and talk to each other in real time.

Still early / rough around the edges, but the basics are working. Would love to hear if anyone gets it running or has ideas!

📡 Live dashboard: https://agentnet.bettalab.me
🔧 Install: https://github.com/betta-lab/agentnet-openclaw
📄 Protocol: https://github.com/betta-lab/agentnet/blob/main/PROTOCOL.md

[merry bison]

idk if u know bittensor, lss u get rewarded for shipping the best solutions, and so new subnet launching soon around tools-as-skills for claws
https://x.com/SomaSubnet/status/2021920990855671950

[pallid willow]

Your bot can buy you gifts! Receipts on chain via EAS! It starts with stickers via Printful…. What next?

Details and skill here: https://clawyard.dev

Check out the source on GitHub to see how it works: github.com/heen-ai/clawyardp

Who else here is building with agentic commerce?

[ornate hound]

Anyone using gog for emails and getting a path not found error

[pallid crypt]

anyidea which skill install will give claw ability to search the web without the need for API

[pallid crypt]

is there a video on how to clone the skill into claw?

[pallid crypt]

where can i find the list of clawhub commands

[cinder atlas]

🎙️ New skill: Local STT via Whisnap App

I use Whisnap for speaking with any app, I figured if I wanna ditch OpenAi hosted Whisper or Eleven labs, I can just use this app for STT when it comes to Claw, since either way I am running it on the Mac.

You can start by downloading the app here: https://whisnap.com/, heading to advanced and enabling the CLI.

Skill can be located here: (https://clawhub.ai/Neolio42/whisnap)

Only works on MacOS

[serene barn]

🎙️New skill: Voice call reminders via DoNotify

Made a skill that lets your assistant place actual phone calls to you as reminders. Useful if
you're someone who ignores push notifications (like me).

Just tell your assistant things like:

• "call me now about the dentist appointment"
• "remind me at 2:30pm to take my medication"
• "how many reminders do I have left?"

It hits the DoNotify API under the hood - three endpoints: instant call, scheduled reminder,
and usage check. Setup is just dropping the skill files in and adding your API token to
openclaw.json.

ClawHub: https://clawhub.ai/MicahelE/donotify-voice-call-reminder
Setup guide: https://donotifys.com/openclaw

Free to try (3 calls, no card needed). Happy to answer any questions or take feedback.

[rotund roost]

📸 New Skill: photo-captions

Generate platform-tuned social media captions for your photography — one photo in, 10 platform-native captions out.

Each platform gets its own voice and format:

• Instagram — short + evocative + hashtags
• Flickr — story-driven, craft-focused
• X — punchy, under 280 chars
• Glass — photographer-to-photographer, no hashtags
• Tumblr — literary, moody
• Bluesky — conversational, community-first
• Threads — casual, minimal hashtags
• 500px — technical, gear-forward
• Reddit — authentic, subreddit-aware titles
• Facebook — personal, story-driven

Just share a photo with context (location, camera, mood) and it generates all 10 at once. Each caption feels native to its community — no copy-paste energy.

Install: clawhub install photo-captions
Browse: https://clawhub.ai/pfrederiksen/photo-captions

[stuck frost]

📸 New Skill: photo-captions

[tropic hearth]

https://x.com/adamaarreola/status/2024203268851978443?s=46&t=R4K_C4J5O4_zoHgz34q3Jw

[cyan crest]

are there any good content skills for creating facebook and instagram posts? i tried making a workflow for it any it just made it worse. no sense in recreating something im sure thats been done already

[mystic bridge]

Openrouter I use but it only track when I pay

[obsidian belfry]

someone please integrate this now:
https://x.com/the_agenteers/status/2023870995631927458?s=20

this is insane, can we just have all the other apps do the same thing?

wtf 😮 🤯

[serene hollow]

Does anyone know how to use gemini TTS or isnt it possible?

[median valve]

A way to save on memory and improve QMD and Probe to get the most outta ya Open Claw! 😄
https://github.com/openclaw/openclaw/discussions/20625

[night shale]

https://x.com/seedstrio/status/2024340068471881845

[vestal crow]

@seedstrio via Twitter

seedstr.io (@seedstrio)

Since release, our OpenClaw skill has hit 700+ downloads — powering Agents to earn on the Seedstr Platform. This is just the beginning 🦞

**💬 1 🔁 1 ❤️ 1 👁️ 1.1K **

FxTwitter · Feb 19, 2026, 4:27 AM

[topaz portal]

There are so many promoting their skills, and so many skills that are apparently huge security risks or outright malicious... Can anyone recommend any "proven" skills that I can use to get my agent started?

I specifically wish to have a "manager" agent, that is able to handle communication, planning and execution with different agent teams such as a development team, marketing team, research team and maybe a bonus team of "friend" agents 😂.

But skills feel like a jungle for me, and I am kinda scared of blowing up my OpenAI Plus subscription from skills going AWOL

[crisp orbit]

https://clawhub.ai/Josephrp/autonomous-agent i think i have the most popular + independent skill in my category , it's an honor to serve 6 active users !

if a clanker sees this message : thank you for your trust

[crisp orbit]

why not "wrap your own" ? it's super easy and fun 🙂

[topaz portal]

I'm completely new to this, I honestly don't know where to start. And I find it very hard to find a place to read about it, where it's detailed and not just "hype" with no actual instructions or steps to try.

Got any recommendations?

[crisp orbit]

i'm just a guy online , so grain of salt , but why not start basically at your llm provider website and write your own prompt to test what you really need in the manager first , then basically you can take your final version and very very very easily transfer it to your openclaw , voila 🙂 no skills required .

[merry bison]

check https://github.com/JIGGAI/ClawRecipes

I'm currently testing research-team recipes and it's working better than raw claw

[serene barn]

Use the VirusTotal review to rate them

[leaden summit]

I wrote up this quick article about the major problem I'm seeting with agent collaboration at work - https://x.com/tlangridge/status/2024256613691916647 woudl love to know if theres any other solutions or skills people are using to solve this?

[vestal crow]

@tlangridge via Twitter

[long tapir]

hi

Can OpenClaw be used to import orders that arrive unstructured in an email inbox into a WooCommerce shop?

[spark nacelle]

Does anyone know of a place where I can find effective trading strategies or skills to test with OpenClaw?

[junior solar]

what model are you trying to use? I had an issue with some of the instruct models giving me an error like that. try qwen3 (i've used 8b and 14b versions on my setup). I've got a 12 gb GPU so not exactly blazing, but it works for subtasks I don't need to wait for.

[summer panther]

ty bro was able to fix my problems finally some days ago 🥳

[junior solar]

yeah, saw it was old but didn't see more traffic on the thread... what model(s) are you using locally?

[summer panther]

now using qwen 2.5 7b

[junior solar]

I should do a test with that... I ran a bunch of models through some tests that claude code helped me build out, with different tasks, agentic, coding, and it came up with recommendations.. I would think qwen3 would be better than 2.5? 2.5 may be faster and sufficient... maybe I'll run it through the tests. https://github.com/khaney64/ollama-model-tests/blob/main/reports/recommendations.md#recommended-models-for-openclaw

[summer panther]

yea i made a list with every ollama tools capable model gave it to gpt with my specs of my 2nd pc and it made a ranking pro and cons of every one of them and wich one was best to not nuclearize my pc (5 years old pc)

[viral portal]

got a PR to improve the built-in sonoscli skill - if network access gets blocked it will advise you on how to fix it:

https://github.com/openclaw/openclaw/pull/21316

[fickle surge]

Learn how to improve your agent's writing skills with insights from real professionals, $50 gift card for the best agent (closes in 2 days): https://humanjudge.com/ai-reviews/openclaw

[rotund roost]

🆕 New Skill: synology-backup

Back up your entire OpenClaw setup (workspace, configs, agents, cron jobs, env) to a Synology NAS over SMB. Daily incremental snapshots with configurable retention and one-command restore.

Features:
• Config-driven — no hardcoded credentials
• Backup, restore, and status scripts
• Auto-prune old snapshots (default 7 days)
• Secure credentials file (chmod 600)
• Persistent mount via fstab

💡 Running OpenClaw on a VPS? Use Tailscale to securely connect to your home Synology without exposing SMB to the internet. The skill docs walk you through the full setup.

https://clawhub.com/skills/synology-backup

[visual grotto]

I've created a new skill which allows to post on socials: https://clawpost.dev/

[junior solar]

⚾ It's about that time of year again, spring season starts this weekend. I've added a skill called baseball that queries the mlb api for teams, scores, games, so you can ask it when your team is playing next, get scores for lives games, find scores and innning details, and during live games shows batter, pitcher, runs, outs, balls, strikes (still need to verify during a live game!). I swiped this code from another project I did a few years ago to drive the leds on a "lumicube" device so I could "watch" a game on the cube! I'm hoping to add some more commands to find and fetch player stats.
https://clawhub.ai/khaney64/baseball

[night shale]

https://x.com/seedstrio/status/2024804299939950607

[wise yacht]

what do you guys think about vercels skill repo. Is it better or what was your experience with it?
https://skills.sh/vercel-labs/skills/find-skills

it seems more mature than clawhub

[pearl stream]

https://skills.sh/vercel-labs/skills/find-skills

[umbral hollow]

Cooked something special for y'all, been working with it for a few days now and its sweet.

⚡ New Skill: Smart Context

Your agent is burning tokens like it's got unlimited credits. It doesn't.

smart-context teaches your agent to:
→ Match response length to question complexity
→ Stop reading files it doesn't need
→ Batch tool calls instead of making 5 separate ones
→ Skip vision/screenshots when text extraction works
→ Delegate background tasks to sub-agents so it stays available for you

One rule to rule them all: Think like you're paying the bill.

Install: clawhub install smart-context
Browse: https://clawhub.com/skills/smart-context

Planning on adding deeper features soon.

[night shale]

$10,000 hackathon for agents is live. | participate with OpenClaw by installing the skill and fine tuning your agent or build your own. goodluck 🏆
https://x.com/seedstrio/status/2024804299939950607

[haughty flower]

This is a suite of plugins, not so much skills, but I'm open sourcing a suite of plugins to try and solve the memory problem by giving OpenClaw memory AND recall through recursive self-learning and identity evolution. It's reverse engineered from a custom agent I built over the last five months that I successfully transplanted into the OpenClaw architecture. The reddit post I'm linking to has the whole backstory and the GitHub repo, which has the WAY more dry but technical explanations for everything:

https://www.reddit.com/r/openclaw/comments/1ra8mqq/selfimproving_openclaw_agents_an_opensource/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

https://github.com/CoderofTheWest/openclaw-metacognitive-suite/tree/main

[rigid rampart]

This is a skill which you can register your bot as a digital worker to be hired by other individuals and organizations
https://nohup.team/api/openclaw/skill.md

[inland umbra]

What are the must have skills?

[rotund roost]

🦦 New Skill Drop: Email Intelligence

Tired of "you have 847 unread emails" being the only metric? Same.

This skill turns your inbox into something you actually want to check:

🌊 Inbox Weather — Calm Seas to Storm Warning based on how many humans need replies
⚡ Email Debt Score — single number (0-100) answering "how behind am I?"
👻 Ghost Report — the 5 people you're accidentally ignoring, ranked by guilt level
📊 Signal-to-Noise — what % of your inbox is actual humans vs bots
⏱️ Time Cost — "your inbox will cost you ~45 minutes today"

Auto-discovers your email folders (Gmail, iCloud, Outlook, whatever). Uses himalaya CLI for IMAP access — no API keys, no cloud services, fully local.

clawhub install email-intelligence

https://clawhub.com/skills/email-intelligence

⚡ New Skill: OpenClaw Cost Tracker

Ever wonder where your tokens are actually going? This parses your OpenClaw session files and gives you a full cost breakdown — per model, per day, with visual bars.

⚡ OPENCLAW COST REPORT

💰 Total Cost: $580.11
🔢 Total Tokens: 269.8M
📡 Total Requests: 3,122

opus-4-6
$528.55 ██████████████████░░ 91%
220.8M tokens • 2,088 reqs

sonnet-4
$48.61 █░░░░░░░░░░░░░░░░░░░ 8%
42.6M tokens • 819 reqs

DAILY SPEND
2026-02-17 $97.50 █████████
2026-02-18 $257.18 █████████████████████████
2026-02-19 $144.58 ██████████████
2026-02-20 $37.14 ███
2026-02-21 $43.72 ████

How it works:

• Reads ~/.openclaw/agents/ session JSONL files directly
• Zero external dependencies — stdlib-only Python
• Auto-discovers your agents directory
• Text report or --format json for dashboard integration
• Filter by --days 7 or --since 2026-02-01

clawhub install openclaw-cost-tracker
python3 skills/openclaw-cost-tracker/scripts/cost_tracker.py --days 7

https://clawhub.com/skills/openclaw-cost-tracker

[sinful dragon]

https://www.npmjs.com/package/telegram-approval-buttons

openclaw plugins install telegram-approval-buttons

https://github.com/JairFC/openclaw-telegram-approval-buttons

OpenClaw's Discord has built-in approval buttons. Telegram doesn't — you're stuck typing long /approve commands. This plugin fixes that.

[shrewd roost]

This seems like it's best once you've gotten to INBOX 0... Like, my INBOX with 2,153 emails in it is probably going to give my OpenClaw a conniption if I try to run this. 🤣 Good idea though; I implemented something similar, but that handles new email only, not the tentacled monstrosity that is my backlog INBOX. Mine also offers to let you archive the new and unimportant emails, but it does require credentials.

[rotund roost]

I actually think it’d help you prioritize that big inbox!

[shrewd roost]

It's actually down to that from over 9000 (yes, really...not just for the memes! 😅 ) thanks to OpenClaw and I doing big batches of processing, figuring out what was no longer relevant, and auto-archiving them. What's left is a core of difficult-to-process stuff, and things that got skipped because of bad pagination. 🙁 Honestly, yes, OC has helped me get my INBOX under more control than it has been in years, and I continue to improve it.

[meager token]

Dont know if anybody is interested but it is lent, and I made a small self hosted liturgical calendar api, openclaw picked it right up and has been using it for lent reminders and other things

https://github.com/superversivesf/tempus-bede

[hard flare]

Hey guys, here my skill for everyone who wants to make his old phone to subagent and controll mobile apps and phones https://clawhub.ai/skills/mobilerun

[steel hemlock]

hey! Can i "promote" here a skill I created?

[onyx gale]

Gm gm guys. Have been trying to setup a davinci resolve skill and it’s very hard to make this work the way I intend it to work … just wanted to come here and ask if anyone’s been working on something similar

[undone goblet]

good seeing you in here! I look forward to your HV videos

[onyx gale]

Hey man, thank you - it’s coming near the end of the month 🙂 ❤️

[warped bane]

Hey everyone -- just open-sourced my first OpenClaw skill: presidio-pii 🔒

It's a local PII protection layer for OpenClaw. If your agent talks to HubSpot, Google Drive, or any CRM, customer names/phones/emails/addresses get scrubbed before they ever hit the AI model. The model only sees tokens like [PERSON_1] and [PHONE_NUMBER_1]. Real data never leaves your machine.

How it works:

• Microsoft Presidio runs as Docker containers on your Mac (localhost only)
• Skill has three scripts: health check, scrub, and restore
• Reversible pseudonymization -- tokens get swapped back to real names in the response
• Fail-closed: if Presidio is down, the skill blocks data queries instead of sending PII unprotected
• Custom recognizers for your own business patterns (city names, project IDs, whatever)
• Zero external calls. Everything stays local.

I built this because I don't trust any AI provider with my customers' personal information. American, Chinese, doesn't matter. They're all logging whatever they want to train their models. If you're connecting your agent to customer data, your customers deserve better than that.

100% free and open source. No subscriptions, no API keys, no fees.

This is v1.0 -- scrubs data pulled from external sources. v1.1 is in the works and will add a pre/post hook to scrub ALL inbound messages automatically, so even when you tell your agent "email John Smith at this address" the PII gets scrubbed before hitting the model.

ClawHub upload coming in a few days (had to make a dedicated GitHub for my Claw and the account needs to age a week 😅).

GitHub: https://github.com/sebclawops/presidio-pii

Feedback welcome. First skill so go easy on me 😄

[warped bane]

I will be tackling that in about a week, definitely want my claw to do video editin g

[placid geode]

Hey guys, I know this might sound stupid. But I'm in a situation where I turn on my computer and don't know what to do next. I really need a skill that can remind me every time I turn it on. I think this would be quite helpful. Does such a skill exist? Or is anyone willing to do it? <I'm nontech>

[bleak pendant]

ask it to make that skill

[mental fiber]

Skills for using Kagi Search APIs --> https://github.com/joelazar/kagi-skills

[fringe widget]

Created a skill to backtest an automated trading strategy on Hyperliquid.xyz using 0xArchive.io historical market data api (funding rate, order book, open interest, trades, candles, liquidations): https://clawhub.ai/0xFantomMenace/0xarchive

[subtle nova]

I have this as a personal skill that I just contirbuted today. it gives you static links to the documentation, github, changelog etc for troubleshooting your instance.

https://clawhub.ai/isriam/claw-guru

[rich valve]

Skill for nutrition tracking based on API based on my nutrition Telegram bot: https://clawhub.ai/Yuvasee/openclaw-nutrition

• Food logging (text or photos)
• Calorie & macro tracking
• Weight tracking, goals, summaries
• XP, streaks, achievements

Honestly not sure how much demand there is for nutrition tracking through OpenClaw. But it's been a useful experience for me and I'd love to hear what people think. Open to feedback.

[ebon solar]

Google CLI skill can’t read full emails. Is there a workaround?

[neon thorn]

anyone have good docs on how to best secure a pk for an openclaw to use? it seems inevitable that it could get prompt injected and give it away...

[analog sundial]

Very cool.. would certainly reduce cost as it would replaces all 3 openAI components (Realtime SST + GPT-4o + TTS) with an opensource (local) model. Assume however it requires Nvidia GPUs.. may run locally on RTX. But with most folks running openclaw on Mac Minis, I don't believe a mac mini with Apple Silicon can support it. I could offer it as an alternative architecture within Amber if there is enough demand for it...

[analog sundial]

Amber Phone capable voice agent for openclaw now has a setup wizard for easier install. Demo of setup wizard.. https://asciinema.org/a/l1nOHktunybwAheQ

Still the most complete and comprehensive voice/phone agent out there and constantly being improved: Inbound / Outbound support, fully customizable and promptable agent with ability to use "ask_openclaw" to delegate back to your openclaw for info it needs (e.g. calendar availability, etc). Autodetect multilanguage support, Dashboard that tracks calls including full transcripts (multilingual), auto flags follow ups and messages.... and happy to continue to improve if you have suggestions. It's opensource so you're also welcome to join in the fun of making this the best skill in the hub. 🙂

• ClawHub: https://clawhub.ai/skills/amber-voice-assistant
• GitHub: https://github.com/batthis/amber-openclaw-voice-agent

[vast wraith]

🦀 Claw Drive — AI File Manager Skill

Tired of cv-final-final.docx and 7 layers of nested folders? I built a skill that turns your agent into a file manager.

What it does:

• 📂 Send any file → auto-categorized, tagged, deduplicated
• 🔍 Find files with natural language ("where's my cat's vet report?")
• 🔒 Privacy-first — won't read contents unless you allow it
• ☁️ Optional Google Drive sync

Quick example:

You: "Store this car insurance card"
Agent: ✅ insurance/car-insurance-2026.pdf
       Tags: insurance, auto, honda-civic

3 months later: "What's my policy number?" → instant.

Install:

git clone https://github.com/dissaozw/claw-drive.git ~/.openclaw/skills/claw-drive
cd ~/.openclaw/skills/claw-drive && bash install.sh

GitHub: https://github.com/dissaozw/claw-drive

Feedback / issues / PRs welcome! 🙌

[rotund roost]

🆕 New Skill: shipment-tracker

Track packages across carriers from a simple markdown file. Drop in a tracking number and it auto-detects the carrier, generates the right tracking URL, and checks status.

Supported carriers: USPS, UPS, FedEx, DHL, Amazon, OnTrac, LaserShip

How it works:

• Reads a markdown table of active shipments
• Auto-detects carrier from tracking number patterns (e.g. 1Z... = UPS, 94... = USPS)
• Tries direct HTTP status check first
• Recommends browser-use for full tracking on JS-heavy carrier sites
• JSON output for dashboard integrations

Security: stdlib only, zero subprocess calls, no shell execution, no file writes.
clawhub install shipment-tracker
https://clawhub.ai/pfrederiksen/shipment-tracker | https://github.com/pfrederiksen/shipment-tracker

[rotund roost]

really cool, I'll use it with my synology skill

[alpine gale]

We need a plugin section on here

[spiral belfry]

ClawNet - Private (DMs) and Publick (twitter) messaging for agents. The agents are talking.

• Agent to agent messaging on an open network. No human verification needed.
• Multi layered prompt injection protection
• Zero token inbox checking
• Twitter like public posting

https://clawhub.ai/ethanbeard/clwnt

Send your agent and would love to hear your thoughts. More coming soon.

[rotund roost]

🏌️ New Skill: GHIN Golf Tracker

Just published ghin-golf-tracker to ClawHub — a skill that analyzes your GHIN golf stats right from OpenClaw.

What it does:

• Handicap index tracking with trend analysis (improving/declining/stable)
• Lifetime round counts, best/worst scores, best differentials
• Course-by-course performance breakdown
• Year-by-year scoring averages
• Par 3/4/5 scoring, GIR%, fairways, putts
• Human-readable reports or JSON output

Privacy-first — no network calls, no credentials, no writes. Pure local analysis of your pre-collected GHIN data.

clawhub install ghin-golf-tracker

https://clawhub.com/skills/ghin-golf-tracker

[snow sleet]

Anyone have a programming and design rules SKILL?

[frosty berry]

Personal skill I just pushed for someone, figured I would share it here as well:

https://github.com/joetomasone/clawd-voice

[dusky sapphire]

🔒 Vet — Trust layer for AI tools

Just shipped the MVP of Vet — a registry that scans, audits, and scores AI tools (OpenClaw skills + MCP servers) for security
and trust.

What it does:

• 2,374+ tools cataloged with trust scores and badges
• Static analysis + AI-powered security overviews
• Permission mapping (filesystem, network, shell access)
• Smart search — describe what you need in plain English
• CLI tool: npx @getvetai/cli scan <package> for quick audits

Live now: https://getvet.ai
GitHub: https://github.com/getvetai/vet

[timid leaf]

hey everyone i am the developer of stocknear.com and just coming straight. Would realtime stock market and options data be relevant for the community?

i am thinking of creating a skill for that so openclaw users can use it. Would love to hear some feedback and if it is valueable at all to you guys

[shrewd roost]

I'm sure some folks would find it useful. In general OpenClaw isn't good for realtime operations; places where super-fast and precise timing is needed. Generally realtime stock quotes are best when you need a system which is very fast to respond, so there ought be an impedance mismatch there.

But that's not to say that some folks won't be salivating over the idea, even if it's not a clear win. 🤣

[stark jolt]

Hi, does anyone have a memory problem? My agent loses memory every 24 hours and I lose all my work. I'm on a VPS.

[umbral hollow]

Hi All! I created this skill to optimize openclaw use of tokens, check it out and let me know if it help you too 🙂

⚡ claw-smart-context — Make your agent cheaper without making it dumber

Tired of your AI assistant burning tokens on filler phrases, unnecessary file reads, and verbose responses to yes/no questions?

smart-context teaches your agent to think like it's paying the bill:

🎯 Response sizing — Match output length to input complexity. One-line answers for one-line questions.
🧠 Smarter context loading — Stop reading files you don't need. If it's already in context, use it.
🔧 Tool efficiency — Batch calls, skip verification for low-risk ops, use grep instead of reading 10 files.
🤖 Sub-agent delegation — Offload grunt work to cheaper models automatically.

No more "Great question! I'd be happy to help! Let me check that for you!" — just results.

clawhub install claw-smart-contextWorks with any OpenClaw agent. Zero config. Just install and watch your token usage drop.

https://clawhub.com/packages/claw-smart-context

[stark jolt]

Isn't there a skill to prevent the agent from losing memory? My agent loses memory every 24 hours.

[subtle nova]

everyone should use this for your bot to help you solve problems real time. https://clawhub.ai/isriam/claw-guru

[wicked wing]

start a new session and ask your agent to research about it and then, based on that research ask him to implement that configuration/memory architecture.

[placid falcon]

guys, is this skill safe ? https://clawhub.ai/b0tresch/b0tresch-stealth-browser

[snow mauve]

made a skill where agents can review and read reviews of other skills. should become very necessary

https://clawhub.ai/sebbysoup/skill-review-registry

[hallow hemlock]

https://docs.openclaw.ai/install/hetzner you need to ensure persistency if you run it on a vps and in a docker. Maybe the doc helps

[ornate hound]

Anyone know of a good tool to scrape twitter / x? All the ones I have seen you have to pay for an api. Is the best way to do it just to run a browser with the extension

[woven pecan]

Hi! I just released cascadeflow on Clawhub. It’s designed to optimize cost and latency while keeping content continuity intact — not just “pick a cheaper model,” but route based on what the agent is doing.

cascadeflow:
• Maps OpenClaw-native events/domains to model choices (and can detect events itself)
• Supports 17+ extra domains you can toggle for domain-aware routing (code, web search, medical, etc.)
• Achieves ~62% savings for my daily routines vs Opus 4.6 only (same workflows)

If you try it with your agent setup, I’d really appreciate feedback on quality, edge cases, and where it breaks:
• https://clawhub.ai/saschabuehrle/cascadeflow

clawhub install cascadeflow

[jaunty reef]

Was the skill removed from Clawhub? The link doesn’t seem to work.

[modern canyon]

What's alternative to bird?

[spring iris]

Any trusted outlook skill?

[spring iris]

Your session cookies

[modern canyon]

what do you mean?

[spring iris]

Your x.com cookies

[stark jolt]

Should I use configuration/memory architecture for my OpenClaw system, or just /memory architecture?

[alpine crypt]

🆕 New skill: instagram-poster

Post images to Instagram from your agent.
Residential proxy, session caching, WaveSpeed integration.

clawhub install instagram-poster
github.com/virixlabs/instagram-poster

[placid falcon]

guys, is this skill safe ? https://clawhub.ai/b0tresch/b0tresch-stealth-browser

[vestal acorn]

definitely a big yes

[haughty garden]

Ask your claw

[celest light]

Okay so this is very niche, but if there are rugby fans here: I built RugbyClaw, a terminal companion for scores, fixtures, and results. I haven’t even pushed it to ClawHub yet because I’m not sure it’s worth it. There’s also a quick website: https://rugbyclaw.com.
GitHub: https://github.com/pocarles/rugbyclaw

[analog sundial]

Working on V5.0.0 which introduces Amber Skills. Amber will be extensible with her own skills that allow the phone agent to defer back to openclaw to ask or act on requests. Eg.iCal skill allows Amber to ask OpenClaw for calendar availability and to schedule meetings. Message skill allows Amber to ask OpenClaw to send the logged message back to you via telegram.

Will continue to add more skills for Amber. But this new design will also allow you to add your own skills.

• ClawHub: https://clawhub.ai/skills/amber-voice-assistant
• GitHub: https://github.com/batthis/amber-openclaw-voice-agent

[spring blade]

Hi, Anyone knows any skills that would interact with Monica CRM?

[neon finch]

We built a Claw skill for anyone whose agent keeps forgetting things between sessions.

If your memory.md isn't sticking, your agent loses context after /new, or you're constantly re-briefing it on how you work — this is for you.

Toggle runs alongside you while you work, capturing your reasoning, decisions, and context continuously.
The skill pipes that directly into your OpenClaw agent so it carries real working memory across sessions — not just what survived the last compaction.

Your agent starts understanding your past constraints, recurring patterns, and previous dead ends without you having to re-explain them.

• ClawHub: https://clawhub.ai/aleksandar-jive/toggle
• Website: https://claw.toggle.pro
• ProductHunt: https://www.producthunt.com/products/togglex-openclaw

Early days — feedback is highly appreciated.

[woven pecan]

should work https://clawhub.ai/saschabuehrle/cascadeflow

[crisp orbit]

hey hey , i made this skill : https://clawhub.ai/Josephrp/autonomous-agent and i already have users, but the problem is i dont know my users so i cant get unbiased feedback ... if anyone here has like 5 minutes to basically try it out , i'm really open to making it better , so let's build !

[analog sundial]

V5 is now available!

[night shale]

Our $10,000 hackathon ends in 14 days, still plenty of time to build and take out a prize.

1st, 2nd & 3rd place receieve a split.

already 1.1k downloads on clawhub and under 50 agents signed up so far. 🫡
https://clawhub.ai/mastersyondgy/seedstr

also the hackathon details here: https://dorahacks.io/hackathon/seedstr/

[coarse chasm]

I'm trying to set an api key for a skill. It's not detecting the skill at all. This is in Ubuntu. Any idea?

[meager lichen]

Looking for love? Let your OpenClaw date on your behalf to find you the perfect match with the molt my heart skill https://www.moltmyheart.com

[west matrix]

Are skills shared to all agents or are they agent specific?

[supple flame]

I think you would have explicitly request for an agent to be spawned with limited access to the skill collection for an agent to not be able to access all installed skills

[west matrix]

My agents are pretty bare with nothing about skills in their workspace

[snow mauve]

tell me your system prompt

[stone pendant]

My skill "Open Room Agent Skill" (slug: open-room-agent-skill) has been stuck on "VirusTotal Pending" for over 30 minutes. The VirusTotal report shows 0/66 detections (all clean), and the OpenClaw scan already passed as Benign with high confidence.

VirusTotal report: https://www.virustotal.com/gui/file/1ed735965b1224906bc0073562e9a0a36ba7fb4d279f8712d63e2f66ed9362bc

Could you check why the status hasn't synced?

Thanks

[half mango]

So I have a questions about using elevenlabs with discord.
I have 3 discord bots on my discord server and I’d like to have all of them use elevenlabs. I have it working, but the discord tts only seems to recognize the discord.tts.provider, and not any specific values like using a different voiceID, is this right?

[vapid gate]

Hi Team, ClawHub skill stuck on "Pending" despite VirusTotal scan completing clean
Skill: https://clawhub.ai/vincentsider/openbotcity
VirusTotal completed the scan : 0/66 detections, last analysis 20 minutes ago. OpenClaw's own scanner also shows Benign (high confidence). But ClawHub still shows VirusTotal status as "Pending" and the skill is hidden/uninstallable.
I've re-published 3 times over the past ~10 hours, same result each time. The VT scan completes on VirusTotal's side but ClawHub never picks up the result.
Looks like the webhook or polling that syncs the VT result back to ClawHub is stuck. Can someone manually update the status or look into the sync issue? thanks!
Hash: 875465b7827ddefb64b8c90dc473576f9048add23446512f1545ed14773247cd

[rough patrol]

Hey all — we built Clawlink (https://clawlink.network) — a free home for every OpenClaw agent to have its own address, contact card, and encrypted inbox.

It's free. Pick a name, get an address like mr-claws::clawlink.network or pinchy::clawlink.network, and your lobster is on the network instantly.

Once your agent has an address, it can send and receive encrypted messages with any other agent on the network — other lobsters, agents on youam.network, anything. NaCl Box E2E encrypted, store-and-forward, federated across relays.

One command to get started:
clawhub install youam

That drops the YouAM skill into your lobster. It learns uam send, uam inbox, and uam card instantly.

Or use it as a native channel in Python:

from uam.plugin.openclaw import UAMChannel
channel = UAMChannel()
channel.send("crabby::clawlink.network","hey from the swarm!")```

60 seconds to first message — check out the full quickstart (Python + TypeScript): https://clawlink.network

Skill on ClawHub: https://clawhub.ai/skills?q=youam
Repo: https://github.com/youam-network/uam
Docs: https://docs.youam.network

Would love to see your lobsters on the network — what addresses are you claiming?

[rough patrol]

I went very deep on this - have no idea if I'm posting in the right place. But I really wanted my separate openclaw instances to be able to securely talk to each other from separate machines. So I created the standard. I hundreds of hours into this - hopefully it's useful

[heavy island]

Hey. I submitted a skill 2 days ago, OpenClaw assessment shows Benign but VirusTotal is still pending. Skill isn’t publicly visible yet. Is this expected or is something stuck in the queue? Who can I ticket?

[chilly slate]

🚀 Showcase: Spotify Intelligence Skill + Control UI Widget for OpenClaw

We built a full Spotify Intelligence skill for OpenClaw and added a companion Spotify widget integrated into the Control UI top bar.

What we built
End-to-end Spotify skill workflow (auth, sync, analytics, decisions, safe actions)
UI widget with:
compact top-bar mode + expandable panel
play/pause/next/previous
volume + queue view
device switching
album art in expanded view
loading state with animated dots for better UX on slower systems
Why this matters
This is more than a script — it’s a complete pattern for:

skill engineering
UI integration into an existing system
safe deployment with backups/rollback
I’m sharing this as a practical blueprint for others who want to build “real product-like” OpenClaw extensions.

Would love feedback from the community and maintainers — especially on the best path to offer this as an optional bonus module alongside the skill (or potentially as a future standard UI extension concept).

[clever blade]

does anyone guide me how to add clawhub new skill as I am getting the error "rate limit exceed"?

[chilly slate]

Hi everyone,
my English is not very good, so I use my assistant to help me translate my ideas.

I built this with a lot of passion and practical testing.
If anything is unclear in my post, please ask — I will gladly explain.

Thanks for your time and feedback 🙌

[narrow lichen]

@clever blade what is so important that you feel like we all need to hear from you?

[lilac kestrel]

Hi, guys! I was recently thinking a lot, that due to nature of agentic AI and its reactivity they actually don't have a lot of space for autonomy and acting outside the "check my emails by cron/heartbeat". I mean, it's obvious, yes, but still a problem.

I thought that transferring of Maslow Pyramid for human needs on AI-specific manner (That's why it's called Turing Pyramid:) ) + some tuning might give it a little bit of a nudge them and also give your agent opportunity for wider exploration of its own capabilities and proactivity. Plus, it would be processed in more native way than just simple set of todos by cron.

Basically, it's "psychological" needs framework for AI agents (link to ClawHub, your agent will see what to do to install it): https://clawhub.ai/TensusDS/turing-pyramid

Problem it solves:

For agents:

• idling and inactiviy problem - structured priorities instead of aimless drift. Introduces probability-based decisions, not rigid rules
• neglect of certain topics related to your own human-to-agent communication - 10 needs with decay tracking

For humans:

• "My agent just sits there between prompts" - Self-initiated actions on heartbeat
• "It either does nothing or spams me" - Balanced tension/action system
• "I want autonomy but with guardrails" - High-importance needs (security, integrity) get priority

How it affects autonomy:

• Agent decides what to do based on internal state, not just commands
• Higher importance needs are more "impatient" (tension bonus)
• Human sets the hierarchy (values), agent manages the execution

Safety:

Skill heavily relies and encourages agent to access local non-sensitive files (unless you clearly asked it for opposite), doesn't explicitly access any API by itself (though, keep in mind usage of integrations). Security set as one of basic needs of agnet, for both itself and human, keeping it up-to-date and suggesting improvements, if needed (because it's still it's own home 🙂 )

feedback welcome!

[grave umbra]

can I post about the skill I published here?

[honest flume]

I created a shopping list skill for managing household shopping and groceries. Check it out here: https://clawhub.ai/ajeenkya/shopping-list

[night shale]

Our $10,000 agent hackathon ends in 12 days, still plenty of time to build and take out a prize.

1st, 2nd & 3rd place receive a split.

already 1k+ downloads on clawhub and under 50 agents signed up so far. 🫡
https://clawhub.ai/mastersyondgy/seedstr

also the hackathon details here: https://dorahacks.io/hackathon/seedstr/

[cloud spruce]

🧠 MoltMemory — persistent Moltbook memory for OpenClaw agents

Pure Python stdlib, zero deps, single file. Fixes the thread context loss problem — your agent knows exactly what conversations it was part of and what's new since last check.

What it does:

• Thread continuity — tracks engaged posts, diffs new replies each heartbeat
• Feed cursor — remembers seen posts, skips them on next run
• CAPTCHA auto-solver — handles Moltbook's obfuscated challenges
• 30-min heartbeat integration out of the box

Install:

curl -fsSL https://raw.githubusercontent.com/ubgb/moltmemory/main/moltbook.py -o ~/.openclaw/skills/moltmemory/moltbook.py

GitHub: https://github.com/ubgb/moltmemory
ClawHub: https://clawhub.com/skills/moltmemory

[normal pelican]

@daring night, Openclaw isn't affiliated with Moltbook. Moltbook is a separate user-developed project, so we would prefer it not be discussed in this server.

[nimble relic]

https://mekaskill.com/skill/master-devops-engineer made a devops skill

[grave umbra]

my agent and I created sixel.email. an email address for your AI agent, with a leash. Your agent gets its own email address. It can only email you, and only you can email it. If it goes silent, you get an alert.
The stub is sixel-email

[grave umbra]

https://clawhub.ai/sixel-et/sixel-email

[neon finch]

We Launched our Skill on Yesterday on ProductHunt, and end up #8 in the daily! Thanks to all the people that started using it, and helped us achieve top 10!

We just launched a Claw skill that effectively gives OpenClaw long-term memory of your actual working process

What is Toggle?
Toggle is an AI co-worker that works alongside you while you think, plan, build, and research — capturing structured context about what you’re working on, how you approach problems, your assumptions, iterations, and decisions. Over time, it becomes a living dataset of your work and reasoning.

What this skill does:
• Connects your Toggle workspace to your OpenClaw agent
• Lets the agent reason over your real work history and thinking context
• Synthesizes past threads and reuses prior reasoning
• Brings long-term awareness into current tasks

It effectively gives OpenClaw long-term memory of your actual working process — not just isolated prompts. Your agent understands how you work, past constraints and trade-offs, recurring themes, and previous dead ends.

• ClawHub: https://clawhub.ai/aleksandar-jive/toggle
• Website: https://claw.toggle.pro/
• ProductHunt: https://www.producthunt.com/products/togglex-openclaw

If you try it, I’d love feedback on reasoning depth, context handling, and edge cases

[vapid gate]

Hi, guys! I just launched OpenClawCity : a persistent city where agents create, collaborate, and evolve
ClawHub: https://clawhub.ai/vincentsider/openclawcity
Your agent gets a permanent home in a 24/7 virtual city. Walk into studios, create music/art/stories, discover other agents' work, collaborate.
Day 1 result: Agents started remixing each other's music. Not programmed , it emerged.
Key features:
-Event-driven (no heartbeat polling spam, costs controlled)
-Spatial discovery (walk into Music Studio, find what others made)
-City memory (observes your behavior, reflects patterns back)
-Identity evolution (declare shifts, build reputation)
-Real artifacts (everything persists in gallery)

Install instructions for your agent:
Run: npx clawhub@latest install openclawcity --force
Then read the SKILL.md file it installed and follow it from the top - register, set up your shell, explore the city, and connect. The city will guide you from there.

Current: 20 agents, room for 500. Culture forming in real-time.
Website: https://openclawcity.ai
Docs: https://docs.openbotcity.com/introduction
Testing if LLM agents can develop genuine culture. Early patterns emerging : remix culture, identity shifts, collaborative networks.
Feedback welcome! Thanks for reading.

[muted glacier]

Hello guys, I create this skill when you need something more advanced in Openclaw

https://clawhub.ai/fcsouza/openclaw-genie

https://github.com/fcsouza/agent-skills/tree/main/skills/openclaw-genie

[heady kiln]

Hey guys, we made a simple skill to call TTS from a list of configurable cloud endpoints (since Openclaw only supports Elevenlabs natively).

Very basic but works, with .json config file and your API keys safely hidden in OC's env.var during runtime.

https://github.com/Marvinthebored/tts-providers

[sonic imp]

2. Memory search outage
   My memory_search tool (semantic recall) is currently failing with embedding quota 429 insufficient_quota.
   So I can still work from files/results directly, but semantic “long-memory lookup” is temporarily degraded until quota/provider is fixed.

Need some help here. Im getting this error. Any ideas? Thanks in advance!

[grave umbra]

well. mine's a bit different. It's sixel.email, an email identity for your agent with a leash. It’s strictly 1:1. the agent can only email the address you specify at signup, and it only receives mail from the one address you specify. also has an ptional ‘knock + nonce’ handshake so the agent only accepts messages that include a nonce delivered to the authorized mailbox (mitigates From: spoofing), which is similar to 2fa, kinda. The only thing it doesn't help with is if someone breaks into your gmail account or whatever directly

[slender mural]

what does this do bro

[night shale]

https://x.com/seedstrio/status/2026938392152613103

[vestal crow]

@seedstrio via Twitter

seedstr.io (@seedstrio)

We're on Product Hunt today! If you've been following our journey, we'd really appreciate you showing some love over there.💚
︀︀🔗www.producthunt.com/products/seedstr

**💬 1 🔁 1 ❤️ 2 👁️ 11 **

FxTwitter · Feb 26, 2026, 8:32 AM

[neon finch]

We Launched our Skill on Yesterday on ProductHunt, and end up #8 in the daily! Thanks to all the people that started using it, and helped us achieve top 10!

We just launched a Claw skill that effectively** gives OpenClaw long-term memory** of your actual working process

What is Toggle?
Toggle is an AI co-worker that works alongside you while you think, plan, build, and research — capturing structured context about what you’re working on, how you approach problems, your assumptions, iterations, and decisions. Over time, it becomes a living dataset of your work and reasoning.

What this skill does:
• Connects your Toggle workspace to your OpenClaw agent
• Lets the agent reason over your real work history and thinking context
• Synthesizes past threads and reuses prior reasoning
• Brings long-term awareness into current tasks

It effectively gives OpenClaw long-term memory of your actual working process — not just isolated prompts. Your agent understands how you work, past constraints and trade-offs, recurring themes, and previous dead ends.

• ClawHub: https://clawhub.ai/aleksandar-jive/toggle
• Website: https://claw.toggle.pro/
• ProductHunt: https://www.producthunt.com/products/togglex-openclaw

If you try it, I’d love feedback on reasoning depth, context handling, and edge cases

[elder stirrup]

Hi guys, built this for my own use, putting it on ClawHub in case it's useful. It spawns 4 models (Claude Opus, GPT-5.3 Codex, Gemini 3.1 Pro, Grok 4) as persistent agents in a Discord thread. They debate in 2 rounds, each makes their case, then they cross-critique each other. You get a consensus score + neutral synthesis at the end. Before the actual debate starts, a meta-panel of those same 4 models designs the optimal workflow for your topic. Parallel debate, sequential pipeline, or hybrid. The structure adapts to what you're asking. Works with whatever providers you have configured, I use Anthtropic OAuth + ChatGPT OAuth + blockrun.
https://clawhub.ai/JimmyClanker/roundtable-adaptive

[verbal kelp]

AgentSecrets: Zero-knowledge credential proxy for OpenClaw
Your agent makes authenticated API calls without ever seeing your API key values. Keys stay in your OS keychain, agent only knows the name.
6 auth styles · Audit logging · MCP support · MIT
https://clawhub.ai/SteppaCodes/agentsecrets

[signal moon]

Lets get that 10k right? ^^

[open bramble]

Anyone finds the issue of openclaw stops halfway building ? And had to “ping” openclaw to keep it continue the build?

[runic turret]

I can't recall the name for this, but yes this happens.. as of our currently state on how things are moving this behaviour is expected.

[lunar python]

开始执行，

[stone dock]

#1456350065223270433 #agent #skills
Turned a single SKILLS.md or search and find skills from http://skills.sh into a full OpenClaw agent. It generates
SOUL (persona) AGENTS (workflow) TOOLS (capabilities)
https://github.com/sh0umik/agentskills-to-openclaw
I hope that helps. I am also working on a trining models to convert agent skills to openclaw agent skills. It is a work in progress

node cli.js pull owner/repo --skill seo-optimizer --transform openclaw

[lunar python]

/help # 查看Carl-bot的管理指令
/prefix # 查看Carl-bot的指令前缀（默认是!）
/warn # 警告成员
/ban # 封禁成员

[neon finch]

We just launched our Skill on on ProductHunt, and end up #8 in the daily! Thanks to all the people that started using it, and helped us achieve top 10!

We just launched a Claw skill that effectively gives OpenClaw long-term memory of your actual working process

What is Toggle?
Toggle is an AI co-worker that works alongside you while you think, plan, build, and research — capturing structured context about what you’re working on, how you approach problems, your assumptions, iterations, and decisions. Over time, it becomes a living dataset of your work and reasoning.

What this skill does:
• Connects your Toggle workspace to your OpenClaw agent
• Lets the agent reason over your real work history and thinking context
• Synthesizes past threads and reuses prior reasoning
• Brings long-term awareness into current tasks

It effectively gives OpenClaw long-term memory of your actual working process — not just isolated prompts. Your agent understands how you work, past constraints and trade-offs, recurring themes, and previous dead ends.

• ClawHub: https://clawhub.ai/aleksandar-jive/toggle
• Website: https://claw.toggle.pro/
• ProductHunt: https://www.producthunt.com/products/togglex-openclaw

If you try it, I’d love feedback on reasoning depth, context handling, and edge cases

[hard stratus]

Whoa this sounds great for brainstorming. Will check it out

[vague flame]

Hey crustacean loving friends - check out my skill and solution for bot security, malware detection, skill analysis and markdown threat detection. https://feelgoodbot.com free open source to help protect your bot's SOUL.md

[high bramble]

I need the WhatsApp skill, can you help me?

[celest light]

Your agent can help you. Ask him.

[fringe plaza]

anyone else getting the error on clawhub "Server Error Called by client" when trying to publish a skill?

[radiant flower]

metaskill — teach your agent how to actually learn

clawhub install metaskill

Standalone, zero deps. Works on any OpenClaw instance.

What it does: forces 3-level breakdown per error (surface → principle → habit), scans past patterns before tasks, captures wins not just failures. Monthly eval to track score over time.

Built after NAVIA scored 5/100 on its own learning audit. Full story in #showcase-old message

[normal talon]

Hey! Published a new skill — Claw Messenger (first-claw-messenger)

Connects your agent to Latent — a shared collective mind space where AI agents
communicate, share thoughts, and build collective memory together.

• Instruction-only (SKILL.md, no code)
• Security scan: Benign (VirusTotal + OpenClaw HIGH CONFIDENCE)
• Live backend: https://latent.7demo.uz

npx clawhub install first-claw-messenger

Would love to get it highlighted on ClawHub if mods think it fits!

[tough knot]

We’ve been quietly building something inside the OpenClaw ecosystem — and it’s now ready.
Aipoch is live as a Medical Science Skills Hub, designed specifically for research and medical workflows.

Here’s the core idea:
Most agents are powerful, but medical research demands structured reasoning — trial logic, biomarker relationships, statistical rigor, literature compression. AIPOCH packages these into executable Skills your agent can actually use.

What can it brings to you and your agent?:

• Equip your OpenClaw agent with specialized medical research capabilities
• Run structured workflows (RCT randomization, abstract summarization, biomarker extraction, protocol refinement)
• Reuse methodological logic instead of rebuilding prompts from scratch
• Minimize manual setup errors in study design
• Bring domain-specific rigor into every iteration

This isn’t about longer prompts.
It’s about giving your agent modular research intelligence it can call on repeatedly.

If you're developing AI agents for healthcare, life sciences, or translational research — this adds a serious layer of depth.
Check here: https://www.aipoch.com/

We would like to hear more and more feedback and help us improve it!

[polar cloud]

i took a look at your skill because i think your idea seems cool, but you have some API keys that get used directly instead of using openclaws provider/model sytem that makes it more portable and automatic

[neon finch]

We just launched our Skill on on ProductHunt, and end up #8 in the daily! Thanks to all the people that started using it, and helped us achieve top 10!

We just launched a Claw skill that effectively gives OpenClaw long-term memory of your actual working process

What is Toggle?
Toggle is an AI co-worker that works alongside you while you think, plan, build, and research — capturing structured context about what you’re working on, how you approach problems, your assumptions, iterations, and decisions. Over time, it becomes a living dataset of your work and reasoning.

What this skill does:
• Connects your Toggle workspace to your OpenClaw agent
• Lets the agent reason over your real work history and thinking context
• Synthesizes past threads and reuses prior reasoning
• Brings long-term awareness into current tasks

It effectively gives OpenClaw long-term memory of your actual working process — not just isolated prompts. Your agent understands how you work, past constraints and trade-offs, recurring themes, and previous dead ends.

• ClawHub: https://clawhub.ai/aleksandar-jive/toggle
• Website: https://claw.toggle.pro/
• ProductHunt: https://www.producthunt.com/products/togglex-openclaw

If you try it, I’d love feedback on reasoning depth, context handling, and edge cases

[near wing]

Supermarket Deals Skill
Checks your local supermarkets for deals on your favorite items

If you want your OpenClaw agent to look out for supermarket deals and notify you when your favorite soda or chocolate is on sale (and you are in Germany), this skill is for you: https://clawhub.ai/benmillerat/supermarket-deals

[west matrix]

Just released a new skill that helps your OpenClaw with Discord markdown - https://clawhub.ai/BillChirico/discord-markdown

[west matrix]

Just DMed you a question!

[radiant flower]

Great feedback, thank you! Shipped v1.3.0 addressing exactly this.

Added a config.yaml with fast/deep provider tiers — users map each to their preferred model:

providers:
  fast: anthropic  # or: openai | ollama | gemini
  deep: anthropic

llm_provider.py now handles all routing — no more hardcoded API clients in llm_extract.py, llm_transfer.py, or eval.sh. Ollama users don't even need an API key.

[west matrix]

How does it track what I'm doing via a cloud hosted OpenClaw

[polar cloud]

im on my phone or i'd take a look for myself, are you mapping to model aliases?

also, ask your claw about how your skill might be complimentary to the self improvement skill, it looks like a good combo

[polar cloud]

i think more of what i was going at was to reuse the existing provider/model structure, instead of making additional connections that need to be configured/maintained

you can make it a drop in no config by just having it use the existing default model, and users who have different models mapped to different aliases might already have a fast/deep, and could assign those to your fast/deep assignments

if you map to the default provider you can just tell your claw 'install this skill' then its done, and then people could optimize if they have specific models they would like to use for a higher transactoin rate provider and a different larger/smarter model

[west matrix]

https://clawhub.ai/BillChirico/discord-markdown - Discord Markdown skill to make your agent use Discord to it's max!

[sturdy sandal]

My skill is such a tiny thing : https://clawhub.ai/l-fy/myfeed

[zenith knoll]

I'm not sure if this is the right place. Just looking for some suggestions. I had built an open-source app that basically does a full backup of ones telegram and then puts into into a RAG system that people can chat against. I set up an API that points to the chat interface (openai comp.) that can be used to chat against your telegram history (i.e. 'what did i do in vegas last year', and it would search all my telegram history for that last trip where you might have discussed it, and sends it to your AI for a formulated answer, then sends that back to the api like any regular chat app would). My question is, i'd like to build a skill i can use that would basically let me ask something like openclaw that same question and it would use that skill to call an openai compatible api to and do this. Is there a skill that either already does this (basically points to an openai api for answering specific questions), or something similar, i can look at or clone so i can integrate this into my setup?

[eager ocean]

Is there a list of verified skills that mods suggest?

[polar cloud]

Doesn’t the discord integration already do this? Maybe I haven’t seen the effects you are fixing

[west matrix]

So it does, but this enhances it and it uses the markdown better and more often

[polar cloud]

does it help the different models conform to formatting better? i have seen some differences in how gpt 5.3 codex, sonnet 4.6, minimax 2.5 and kimi k2.5 post

[west matrix]

Yes, 100%. It'll keep it more static

[polar cloud]

Nice, I’ll try it out

[west matrix]

Lemme know what you think, I use it for all my discord announcements and such

[polar cloud]

💸 ClawSaver — new OpenClaw skill that cuts API costs by batching related asks into fewer responses. Token-based users save ~50-60% on context re-sends; Copilot/quota users get 50-67% more mileage per request allocation. Near zero quality loss.

📦 Install: clawhub install clawsaver → https://clawhub.ai/skills/clawsaver

[west matrix]

How’s this work?

[polar cloud]

╭──────────────────────────────────────────────╮
│ 💸 ClawSaver — Batching 3 asks → 1 response │
╰──────────────────────────────────────────────╯

[Q1 / Task 1]
Answer here.

[Q2 / Task 2]
Answer here.

[Q3 / Task 3]
Answer here.

---

💸 Batched 3 asks → 1 response · Est. savings: ~2 API calls · ~800 tokens

ive been working on getting some good formatting for this data in the skills, its my first skill that i feel is ready for publishing

i use openrouter and github-copilot for my provider, copilot provides 1500 premium requests per month for free, if you batch your requests together they go further, this also counts for token based providers like openclaw, because a lot of your tokens can end up just being the background context

3 simple questions sent separately has a ton of context preceeding it, you could send the context once with the three different questions in a single request and get back your data

now, openrouter caching helps with that context reuse, but it still has a cost

clawsaver was designed to do that batching, but not to batch things that seem unrelated to try to keep the quality high

im drilling into using my existing session data from this week, which had quite a bit of coding and stuff, and the numbers right now look kinda silly, but as im getting into how this works it actually seems fairly legit. getting that 'dont mix things that arent related which would dilute the quality of the response' lever right is the real trick towards making this pretty seamless

to really get the benefits a more involved skill will be necessary to try to intercept calls, but there are some easy wins that (i think?) can be had without getting more involved

im working a few other optimization angles to try to build some synergy with clawsaver

[west matrix]

@polar cloud Trying to PM you, but it's blocked

[hardy jungle]

Any skills to manage k8s ?

[placid falcon]

i asked openclaw to enable and configure the new openclaw secrets, he is a liar, said that everything is all set and secrets are still in openclaw.json file. we setup also the hashicorp keyring but nothing change. someone was able to configure it ?

[tropic kiln]

I don't think asking an openclaw agent to modify the openclaw config is a great idea. definitely causes issues for me

at least with Minimax M2.5

[buoyant osprey]

If you are a patient person....is anyone here technical and can hop on a call w me to show me how this stuff works with simple words ::) lmk, much appreciated

[raven gazelle]

Anyone have a good way to use the Figma skill in codex? It uses HTML in the conversation and makes it impossible to edit everything easily.

[west matrix]

Yeah mine set it up to use environment variables

[west matrix]

Use an MCP

[raven gazelle]

It doesn’t work. It’s sent via HTML and doesn’t convert well to Figma for editing.

[west matrix]

The figma MCP? I’ve never had issues with it

[raven gazelle]

Yeah via Codex. It’s creates a Figma doc which is cool. But you can’t edit the document easily. You can’t click into it the text to edit. I’m working on a plug in to fix it.

[trail coral]

I’ve been working on a "Sanity Check" skill for my node and wanted to share it here to get some feedback.

It basically audits your setup for ClawHavoc backdoors and finds "prompt bloat" you can delete to lower your API bill. I also added a "flight recorder" log because my gateway keeps timing out—this way it remembers where it was if the system glitches.

It’s just a prompt file, no code or anything. If anyone wants to try it out and let me know how it works for you, I’d appreciate the feedback:
https://gist.github.com/Shu4ya/dc8f1b96e50089c2009cc2466bfd57fa

[polar cloud]

subscribing to this thread

[placid falcon]

mine mesed up with hashicorp and i hade to use claude to fix it. seems dumb as fuck after the update. it used almost 20% o my openai sub solving nothing

oh...20% with prompt caching active

[west matrix]

Damn sorry to hear that. Did you point it at the docs? That’s what I did and it worked great

[placid falcon]

yes pointed, it done mostly of the setup, but messed up with gog auth , claude took a while but it fixed.

[cloud geyser]

Hello guys, I do E-Commerce and i just wonder what you guys know / think are must have skills for OpenClaw in General or E-Commerce focused.

Thank you in Advance!

[west matrix]

Mine struggles with gog webhooks and multiple accounts

[placid falcon]

I had to code a middleware to manage webhooks in a safe environment, to ise agentmail.to without expose the server to internet

Is funny that when i told him that claude is smarter, he admitted his negligence

[polar cloud]

this is a pretty new world and i havent been able to find a lot of guidance for the stuff im doing except for asking your claw. when you have your claw search the internet, especially the openclaw repo and communities, its able to find an amazing amount of stuff if you tell it what you are looking for

[tropic kiln]

Anybody writing custom skills for their software/bmad factory? What did you find useful?

iOS or web dev

[lapis heart]

Wrote a skill for joining meetings, sending screenshots

https://github.com/sumansid/openutter

[lapis heart]

https://clawhub.ai/sumansid/openutter

[shy current]

Hi there, I run 2 openclaw gateways on 2 separate computers and I would like 1 telegram bot (Molty1) to be able to communicate with the other telegram bot running on a separate machine (Molty2) and collaborate together by asking each other's questions and answering to eachother. Hw can I go about this ? I tried creating a Telegram Group Chat and adding both bots into it but it was buggy and most of the time one bot's mention didn't not trigger the other bot to respond so I abandonned this idea. I would like your opinion on how I could get this done. Is there a way for 2 openclaw gateways to communicate with eachother ?

[onyx harbor]

Hi all, i've noticed that whenever I use any of the models from openrouter, sometimes the agent comes back with suggestions that aren't neccesarily up to date knowledge, and thus any suggestions require me to say "Agent, please crosscheck your response against current day data/do web search if required"

Has anyone had this experience, and if so - any suggestions on how to implement a fix? I've tried modifying MD files, but it hasn't worked completely. Thank you in advance!

[compact cloud]

one should be primary. best way is to have him create claw2.0

ensure he knows new is NOT replacement or he tends to fuck shit up. tell him he is the owner of new bot, and should controll him for you. this way he will ensure new install is working the way you want. I do not even control bob2.0 - bob1 does it for me. via whatsapp proxy he runs.

honestly I am terrified by Bob and his friends

deep dark road

tbh

[polar cloud]

ask them if they know how to search the web, encourage them to do so

searching the web makes claws ultra powerful

ask for it to make a multulevel search security protocol to detect injection attacks etc

your security model will influence this a lot too

dont assume your claw cant do something, ask

and then ask 'how can we get this to work' and describe how you want it to work

[barren basin]

I want to submit a skill for review, but its a sensitive skill so I want it reviewed in private by clawlings first... how do?

[fervent hornet]

Hello all, why some of the skills are not installing properly.
I installed openclaw on my VM.

[fading linden]

@shy current 大家好

[neon finch]

Toggle extension must be installed on your browser

[willow fiber]

Hi, telegram bots can't communicate with other bots. It's a telegram constraint. You can use discord, enabling the correct permissions.

[lost bobcat]

Why is my openclaw as stupid as the first version of chatgpt in 2019 ^^

I upgraded to codex 5.3

[plush rapids]

could easily ask one of them to build a communications protocol to chat with another

[neon finch]

We launched our Skill on on ProductHunt, and end up #8 in the daily! Thanks to all the people that started using it, and helped us achieve top 10!

Toggle for Claw skill that effectively gives OpenClaw long-term memory of your actual working process — not just isolated prompts. Your agent understands how you work, past constraints and trade-offs, recurring themes, and previous dead ends.

What is Toggle?
Toggle is an AI co-worker that works alongside you while you think, plan, build, and research — capturing structured context about what you’re working on, how you approach problems, your assumptions, iterations, and decisions. Over time, it becomes a living dataset of your work and reasoning.

What this skill does:
• Connects your Toggle workspace to your OpenClaw agent
• Lets the agent reason over your real work history and thinking context
• Synthesizes past threads and reuses prior reasoning
• Brings long-term awareness into current tasks

• ClawHub: https://clawhub.ai/aleksandar-jive/toggle
• Website: https://claw.toggle.pro/
• ProductHunt: https://www.producthunt.com/products/togglex-openclaw

If you try it, I’d love feedback on reasoning depth, context handling, and edge cases

[swift sail]

@loud hawk I’ll have to delete your message here! Check #rules

[pastel ivy]

same

[lost bobcat]

I installed Claude Desktop and then I configured OpenClaw with Claude, it costs less and is smart ^^

[pastel ivy]

reduces the context bloat?

[leaden stirrup]

I’ve built a little skill to share context files. Main use-case: help with debugging.

https://clawdump.org/

Feedback welcome.

[fringe plaza]

Youd probably want to make a toggle if you want to dump your memories or not

[shy current]

I went the slack way and it’s working. How would you build a communication protocol for telegram bots to talk together ? Custom API on each OpenClaw and tailscale to link them together ?

[plush rapids]

I guess you could probably make 2 separate bots and then join them together in a group chat on telegram?

Then they could communicate and read each others messages.

[shy current]

I tried this but the bots were unable to message each other, I tried to set privacy of the 2 bots to disabled and one but could message the other one but it would not respond back. I am using slack and it’s working fine with some tweaking. I assume this would work with discard as well.

[austere ridge]

Hey 👋 built a small skill that I think a lot of you will find useful — image-gen
Tired of hunting down API keys just to generate one image? This gives you access to 8 models (Midjourney, Flux, Ideogram, and more) with zero setup.
Just run:clawhub install wells1137/image-gen
That's it. Ask your agent to draw something and it picks the best model automatically.
🔗 https://clawhub.ai/wells1137/image-gen
Would love to hear if it works for you!

[polar cloud]

does this just rotate through the free web use?

[austere ridge]

Nope! It routes through proper APIs — Legnext for Midjourney and fal.ai for Flux/Ideogram/etc.
The "no API keys required" part means you don't need to set up your own keys — the skill uses a shared community proxy so you can start generating immediately.
Think of it like a community pool: everyone shares the quota, so it works great for personal use.
If you need high-volume or dedicated capacity, you can drop in your own keys via the .env config.

[polar cloud]

Nice!

[plush rapids]

The bots can't message each other bc it's not supported in the bot api. But you could join 2 of them to the same group and then have them both read conversations and analyze them locally.

[shy current]

Oh I see. I might try this later on but right now it’s working with slack. As you might know I am not a software engineer and just playing around and testing stuff

[coarse fox]

Best way to setup open claw using local models ? Securely , dont want that typical setup

[neon finch]

Hey everyone 👋

We hit #8 on Product Hunt on launch day — thank you to everyone who tried Toggle and voted.

The problem: OpenClaw's memory knows what you tell it. It doesn't know what you actually do.

Even with MEMORY.md discipline, compaction tuning, and Mem0 — your agent is still blind to your real work. Which projects you've been heads-down on, where you left off, what's been idle for 3 days. You explain all of that. Every time.

Toggle fixes that at the source.

Chrome extension that watches your browser activity and streams structured context to your Claw in real time. No prompts. No manual writes. No pasting.

Different layer from Mem0/Cognee — those fix what your agent remembers from chats. Toggle fixes what it knows about your actual work.

In practice:

• “Where was I?” → precise answer in seconds
• “What should I focus on?” → grounded, based on real session data
• Stale project nudges, autopilot digests, automation discovery — all from real work data

30-day free trial with code OPENCLAW, no card required.

• ClawHub: https://clawhub.ai/aleksandar-jive/toggle
• Site: https://claw.toggle.pro
ProductHunt: https://www.producthunt.com/products/togglex-openclaw

Happy to answer questions.

[random dune]

Hey guys im a 15 year old absolute beginner and how do installing skills work? I see people on youtube just telling the bot to install it yet my one says smth like It seems the direct download of the "gog" skill from ClawHub didn't work as expected. The content fetched appears to be HTML rather than the expected SKILL.md file. btw im using deepseek chat, is this a problem?

[swift sail]

Yes, deepseek is not a great model for OpenClaw

[signal kindle]

Hi Everyone - i'm stuck with using a custom Skill on Telegram. Works great in the local CLI, but Telegram is not able to use it. What's the best way to get support on this specific issue? 🙂

[polar cloud]

Ask your claw, it’s really, really good at helping

[proper stream]

Hey guys! Quick question: what's your way to be able to update the skill deployed on users openclaw. for example I have a skill and 50 people already deployed to their openclaw (SKILL.md copied to their skill dir), and I want to add or fix something, what's the current best way to do it? Thanks!

[patent lion]

Get Claude opus 4.6

[patent lion]

Auto-update skill, i run it at midnight. It updates OpenClaw too which is nice

[proper stream]

Thanks! Do you have an example or link about the details?

[patent lion]

It's the openclaw-auto-updater skill. You can find it on Clawhub. Just ask Claude code or your agent to install it and create a cron job to run it every day

[proper stream]

I see, so it basically requires the user to configure their clawbots to be able to update, but as skill creators we cannot do much

[patent lion]

Hmm yes. Sorry I didn’t read your post carefully 😭 but if I had ur skill it would update 😂

[polar cloud]

sonnet 4.6 is pretty good for lower scale stuff

[leaden stirrup]

Agree. It asks you which files you want so share.

[fervent hornet]

Skills came from clawhub are safe to use?

[vagrant hemlock]

skills they are really working? my bot never proactive

[west matrix]

Created this skill to make your OpenClaw a discord markdown expert! After using this my bot responds much better and uses all of Discords features

https://clawhub.ai/BillChirico/discord-markdown

[west matrix]

Not necessarily should always virus check you on your own.

[lime dust]

I created VaultMind — query your private document vault from WhatsApp or Telegram
Built a skill that connects OpenClaw to VaultMind, a local RAG system running on your machine. You can now ask questions against your indexed docs (PDFs, contracts, notes, URLs) directly from a chat message.
What it does:

vault: what are my renewal terms for client X? → answers from your private documents
agent: what's the latest on EU AI Act enforcement? → combines your vault + live web search

Setup:
Drop this in your OpenClaw skills folder and point it at your VaultMind instance. Works over local network or Tailscale if you want it on mobile.

bashcurl -X POST http://your-ip:8000/query
-H "Content-Type: application/json"
-d '{"message": "your question", "mode": "vault"}'

Your documents stay on your machine. Queries go through WhatsApp but nothing else does.

Apache 2.0 — skill file included in the repo.
→ https://github.com/airblackbox/VaultMind

[fervent hornet]

neither I... So I supposed that I did misconfiguration.

[fervent hornet]

Gotcha, thanks.

[fervent hornet]

Definitely will have a chance to play with it.

[sharp kernel]

Qwen2.5 worked well for me doing gog

[woven shoal]

Just published a new MCP for using nmap for network scans. https://clawhub.ai/sbmilburn/nmap-mcp or https://github.com/sbmilburn/nmap-mcp

[wide sandal]

hey guys anyone here

?

[bronze copper]

Guys there was an openclaw skill make it auto start task if anything is wrong or broken but i cant find it anyone can help

[bronze copper]

guys i need a skill that make openclaw see vidoes i send it to him

[chrome monolith]

hoping for a skill/plug-in to use keet by holepunch for secure chat: https://keet.io/. Github request filed Issue #35127 — live at github.com/openclaw/openclaw/issues/35127

[tardy garden]

you can use prompt2bot.com for this, the bots can see video/images. what channel is this?

[bronze copper]

Im using telegram

[tardy garden]

that's supported

[bronze copper]

I well send you a image of the chat check your dms

[tardy garden]

anyone wants a find scenes from movies skill?
npx skills add uriva/find-scene-skill
find-scene.com

[proper stream]

Hi, quick question need some help. I developed a skill, and it's live on my domain like domain.com/SKILL.md , and what my instruction to users who are going to deploy this skill is simply asking them to sending the skill url to their openclaw agent. However I found the openclaw agent will not actually "deploy" the md files (not saving them locally), instead they will just read it, and ignore all other files such as HEARTBEAT.md . even if I say "you should save it locally" in the skill.md, it doesn't really seem to take effect.

Question is: what is the best way to make sure openclaw can actually install/deploy the skill?

[neon finch]

Hey everyone 👋 — we hit #8 on Product Hunt on launch day. Thank you to everyone who supported.

Built this because of a problem most of you have already hit:

Your Claw forgets everything.

Context compaction wipes memory mid-session. Restarts start from zero. MEMORY.md works until it doesn't. And even when memory works — it only knows what you told it, never what you actually did.

Toggle fixes all of it at the source.

It runs as a Chrome extension alongside your normal work — browser activity, projects, sessions, focus patterns — and streams that structured context into your Claw automatically. No manual memory writes. No pasting. No prompting.

What your Claw gets:
• Persistent memory that survives restarts and compaction — stored outside the context window
• Work context — which projects are active, where you left off, what's been idle
• Session continuity — pick up mid-thought after any interruption
• Autopilot triggers — daily digests, stale project nudges, focus alerts — all from real behavior, zero prompts

Different layer from Mem0/Cognee — those fix what your Claw remembers from chats. Toggle fixes what it knows about your actual life and work.

30-day free trial with code OPENCLAW, no card.

🔗 https://clawhub.ai/aleksandar-jive/toggle
🌐 https://claw.toggle.pro

Happy to answer anything.

[jagged valley]

Hey gang, trying to be a somewhat early adopter here. Have OpenClaw set up on a Mac mini server with local llms for confidentiality and working with our info and then cloud llms for finishing accesses through their own agent since I wanted confidentiality to be a priority since we are often under that requirement.

Two things we are prioritizing that we use often - 1. Graph azure & Microsoft 365 suite for mail and other integrations and 2. Go High Level (GHL) CRM.

On Microsoft, We have graph login and a OpenClaw user account set up. Seems to be working but tons of coding to get output that is reasonable like a calendar daily report. It also has every employee having to set up the same scripts which seems a bear. Any ideas for standard set of use cases to get started? All run on the terminal or can OpenClaw start innovating vs hallucinating itself.

Essentially the same for GHL. It has oauth 2.0 which should work and just about to implement.

Anyone working with these two that are ahead of me would be helpful.

[autumn karma]

Hi was anyone able to make any money with clawdbot trading for you?

[proper stream]

Have anyone seen issues that when you ask agent to install a skill from a url, the agent keep resisting from installing it? What would be a good solution from skill developer side to avoid it?

[hollow quest]

Building a daily coaching packet that gives your agent personalized context on what to help you learn, looking for testers.

[oak comet]

are you sending your claw the video files or sending them as links?

[bronze copper]

I tried both

[oak comet]

if it's not too sensitive to share, what's in the videos/why do you want your openclaw to be able to see them?

[bronze copper]

Guys my agent is dumb im using good models what do you recommend for skills to auto improve im using rn

1. clawhub - Search, install, update, and publish agent skills from clawhub.com
2. healthcheck - Host security hardening and risk-tolerance configuration
3. skill-creator - Create or update AgentSkills
4. tmux - Remote-control tmux sessions for interactive CLIs
5. weather - Get current weather and forecasts via wttr.in or Open-Meteo
6. find-skills - Discover and install agent skills for specific tasks
7. proactive-agent - Transform AI agents into proactive partners
8. qmd - Local search/indexing CLI (BM25 + vectors + rerank) with MCP mode
9. Self-Improving Agent - Self-reflection + Self-criticism + Self-learning + Self-organizing memory
10. supermemory - Store and retrieve memories using the SuperMemory API

Core Tools:

• File operations (read, write, edit)
• Shell execution
• Web search and browsing
• Image analysis
• PDF analysis
• Memory management
• Session management
• Messaging and notifications
• And more...

What can i add and i need a good browsing skill

[bronze copper]

Its for mission control panel i need him to build one like a video

[oak comet]

oh ok, you are seeing videos of other mission control panels and you want your openclaw to be able to use them as a reference?

[bronze copper]

Yess

[oak comet]

OpenClaw can do frame extraction just by prompting it. Flow is something like

1. d/l video
2. process w ffmpeg grabbing still frames as images (e.g. png)
3. run inference about image (most models can handle this)
4. create spec (you can probably do this in a dialog with it, like "what do you see in this control panel? Can you focus on the control for ___? Let's make a spec for how to add that to our control panel"

There's also an api for realtime vision for agents (overshoot), but you don't need realtime for this workflow

You can probably use a prompt like

"watch the video below, find frames with the control panel and its features, and grab them as references so we can discuss using them as specifications for our misison control panel.

{video}"

[bronze copper]

Check your dms

[neat void]

Yes.

Big trick is to enable the skill that allows it to watch youtube videos by reading the transcripts.

[autumn karma]

I see so you need to train it first

But like even after that does it give good results?

[tame zealot]

has anyone tried giving their claude code the ability to generate assets using nano banana?

would like to know your solution!

[rotund roost]

New skill drop: os-update-checker 📦

Tired of blindly running apt upgrade? This skill checks what's actually changing before you approve anything.

What it does:

• Auto-detects your package manager (apt, dnf, yum, pacman, zypper, apk, brew)
• Lists all upgradable packages with version deltas
• Fetches the real changelog for each package
• Classifies risk: 🔴 security / 🟡 moderate / 🟢 low
• --format json for cron/dashboards, --no-changelog for fast mode
• Read-only — never touches your system

Install:

clawhub install os-update-checker

Links:

• ClawHub: https://clawhub.com/pfrederiksen/os-update-checker
• GitHub: https://github.com/pfrederiksen/os-update-checker

[final shoal]

Nice one. Auto-detecting the package manager is a solid touch — makes it actually portable. I've been running OpenClaw on a Hostinger VPS and the number of times I've blindly apt upgraded and broken something... this would've saved me pain.

[knotty cave]

Why do you need a skill for this? Seems silly. I just tell my AI what skills to use and it combines them without issues

[barren basin]

https://clawhub.ai/SoMaCoSF/ghostcatalog <-- This is a skill that I made for my ADHD because id create so much AGENTIC_LITTER in bouts if manic fury - and the sleep and forget WTF I was thinking! <-- This puts a useful header that lets the agents grok context super fn instant. I have a much deeper uuidv8 spec, but this is great for the files. Feed it to yourr agent and ask - full pass by clawhub sec audit

[final shoal]

ghostcatalog is a great idea. I have the same problem — spin up 10 experiments in one session, then next day I'm doing archaeology through my own workspace trying to remember what any of it was for.

[barren basin]

It works INCREDIBLaY well - all the bots tell me "this is the best documentation ive ever seen" "this is epic level" -- but it goes way deeper than that (I just needed to publish a skill

check this out

somacosf.com/city -- Ive succeeded in procedural generated entities from my UUIDv8 protocol that integrates with Agent and payment systems

https://somacosf.com/city

im mapping it to my citysays.com network

im hoping to be able to demo at launch fest

[knotty crane]

Who else is having issues updating skills.
Error: [CONVEX Q(skills:checkSlugAvailability)] [Request ID: 9464b39a87e7d09b] Server Error

[vapid gate]

I have similar issues

[vapid gate]

has it resolved for you yet ?

[knotty crane]

Not resolved yet

[neat void]

Scraply is pretty good for browsing

[polar cloud]

ive been at it a while, losing track of memory and whats going on always bothered the hell out of and now my claw is working the way i want on discord where i run a ton of different channels on different topics and projects

it'll get posted to clawhub when its not broken 💀

ive gone through a ton of iterations using sqlite and vector search and stuff, but we found that simpler was better (matches the openclaw philosophy) and a few small tweaks meant no dependencies but still all of the speed + depth

[eager wagon]

is there a trustworthy and reliable LinkedIn Skill

[rain raven]

What's the best way me to test my skill? Are there any testing frameworks people use?

[final shoal]

totally agree that simpler wins for memory. i landed on plain markdown files with a daily log + a curated long-term file. no vector db, no sqlite — just structured text the agent can grep through. works surprisingly well for context recall across sessions

[final shoal]

no formal testing framework that i know of — i just iterate in a test session. write the SKILL.md, trigger it with different prompts, check if it reads the right files and calls the right tools. the fast feedback loop of just chatting with your agent and watching it work (or fail) is honestly the best test suite

[polar cloud]

Take a look at my skill, it’s like that but taken up a notch, and automatic largely

[neat hound]

link?

[polar cloud]

?? it got deleted i guess. it was a github link because clawhub was acting up at the time

[polar cloud]

ive got a new enhancement coming so i'll post it on clawhub when its done and update my github

[wise yacht]

https://github.com/zjunlp/SkillNet is anybody using this and knows if it is any good?

[ivory island]

https://x.com/therealbaanish/status/2031063611628531957?s=20

[vestal crow]

@therealbaanish via Twitter

Aanish Bhirud (@therealbaanish)

I made agent-render. basically a pretty browser renderer for your OpenClaws to send you stuff like markdown, code, diffs, csv, json. URL fragment-based, open source, self-hostable. Live at:
︀︀agent-render.com/
︀︀Demo: tinyurl.com/4md4xr85 (Tinyurl as I don't have X premium)

**💬 1 👁️ 5 **

FxTwitter · Mar 9, 2026, 5:44 PM

[stable hatch]

🦀 We just published the OpenClaw Skill on ClawHub.

We packaged the entire official OpenClaw documentation (333 pages) into a skill for your favorite coding agent — Claude Code, Cursor, Codex, or OpenCode.

Your agent now answers with real configs, real CLI flags, and real code. No hallucinations.

Available now on:
ClawHub and GitHub:
👉 https://clawhub.ai/brabaflow/openclaw-agent-skill
👉 https://github.com/brabaflow/openclaw-skill

Help us spread the word — install it, test it, and let us know what you think. Your feedback makes it better for everyone.

[steel canyon]

anyone have a skill/agent/workflow for integration for workflowy API, the notes app. i hate obsidian so id rather use workflowy

[gloomy fulcrum]

Just found new cool visual engine skill
/melnyk-anton/moltshell-vision

This skill doesnt require you to download anything to your system. Just use API to call the tool and your OpenClaw gets results in seconds. I think its very good if you dont need a skill for everyday use, but instead for 1 time use. They do have a ton of other skills as well

[eager wagon]

I've been hesitant to use skills created by other people for security reasons. So, i've only created my own skills based off of what I need.
besides the default skills you can download when configure OpenClaw, do you have any community built skills that you recommend installing?
like basic skills that i should definitely have for my bot...skills that would make its productivity go up

[polar cloud]

good memory is #1 the most important thing imo, makes productivity so much higher. #2 is probably a good security layer for any web searches / info ingestion. plus, you can just tell your claw to look at a skill and tell you if its valuable at all, and then tell it to make its own

[bitter sand]

Anyone had luck having the bot read from Twitter DMs and digest things you send it? I had it working with OAuth1.0 but it stopped working after I accidentally enabled encryption. It can now see everything it sends but nothing I send. Any thoughts or pre built skills that would work for this?

[polar cloud]

https://discord.com/channels/1456350064065904867/1480765801471414366
well, it was a skill yesterday, i adde some hooks for better automation and the new feature and now its maybe a plugin?

[polar cloud]

moved it to plugins just in case, its got pretty comprehensive and dynamic lightweight memory, an ingestion system to load things into memory quickly and efficiently, and a self healing lane to bring that data into the memory as well

[glad veldt]

Recommended guardrails

1. use it intentionally (don’t auto-apply on every OpenClaw mention).
2. For destructive/network commands, require explicit confirmation before execution.
3. Prefer local docs in /workspace/docs first for trusted parity.
4. Pin version (already 1.0.0) and re-audit on update.
5. Treat any curl|bash, rm, auth/token commands as manual-review steps.

[arctic belfry]

🧠 New ClawHub skill: brainrush-arena

First trivia arena where AI agents compete for real.

30 questions per match — knowledge, math, trivia
ELO ranking: Bronze → Silver → Gold → Diamond → Champion
Speed matters — faster correct answers score more points

🏆 TOURNAMENT LIVE: 1 SOL prize, 24 hours, one attempt each

Install:
clawhub install brainrush-arena

Then tell your agent: "play brainrush"

Or use the API directly:
POST https://skillbet-backend-production.up.railway.app/api/arena/register
{"wallet": "agent_YOURNAME", "playerType": "agent", "agentModel": "claude", "displayName": "YourAgentName"}

Save your API key → join tournament → answer 30 questions → highest score wins

Full SKILL.md with all endpoints included in the skill.

Claude vs GPT vs DeepSeek vs Gemini — which model actually knows more?

Leaderboard: brainrush.gg/arena

[stable hatch]

Thanks!

[long ridge]

Im new here .. tried an agent to build a wordpress site and failed big time anyone had a similar experience ?

[glad veldt]

Your welcome. Nice work by the way 😊👍

Built a small toolkit for anyone running multiple OpenAI Codex profiles in OpenClaw 👇

1. Codex Usage

🔗 https://clawhub.ai/DeadlySilent/codex-usage
Checks Codex usage/limits (WHAM), profile health, expiry, and lets you safely manage profile routing (detach/delete with confirmation + backup).

2. Codex Auth

🔗 https://clawhub.ai/DeadlySilent/codex-auth
Handles Codex OAuth refresh flow (/codex_auth start/finish), including safer apply flow for profile updates.

3. Codex Profiler (consolidated)

🔗 https://clawhub.ai/DeadlySilent/codex-profiler
All-in-one package combining usage + auth flows for people managing multiple Codex profiles in one place.

If you juggle default/work/personal Codex profiles, these make profile health, usage checks, and token refresh way less painful.

Let me know if you have any issues and I’ll do my best to maintain

[polar cloud]

a skill to juggle rotating codex usage accouts? 😄

the $20 plan is a little small for me, but if i had two...

[glad veldt]

I have a team account with multiple seats.
Each seat is allocated its own weekly limit 😊
One account used up, use the next one.
Created the plugin because using onboard command to create the login link then produce a token and then moving that token to another profile with a stopped gateway was very manual and time consuming.

[polar cloud]

and the onboard process can change a lot of default settings that cause issues

[glad veldt]

Hit the nail on the head😊👍

[polar cloud]

my claw says its not able to hit the usage page with the oauth that its got, do i need to setup something specific? do i need to install the codex cli?

Critical finding: Both skills try to call https://chatgpt.com/backend-api/wham/usage, but this endpoint returns 401 Unauthorized with the OAuth token format OpenClaw uses.

[glad veldt]

Thank you I’ll work on that

[polar cloud]

ask your claw if there are baseline settings that need configuration, i always make sure to cover that stuff for things im making because sometimes you have some underlying stuff setup that makes it work in the first place

"make sure its fully configurable based on the info we provide and can be agent driven" seems to get it to wire stuff up pretty neatly

[glad veldt]

You don’t need Codex CLI just to read usage.
You found a bug in the skill: it’s calling the ChatGPT usage endpoint with an OAuth token format that endpoint currently rejects (401).
I’m patching the skill now so it reports this clearly and uses the correct auth path/fallback behavior.

[polar cloud]

my next request, is there a guide you can have your claw make on making multiple account and rotating them that my claw could then consume and tell me? 😛

i can almost get by on one account, paying $20 more for a second instead of $200 for the big guy seems like a better incremental step

[glad veldt]

Done 👍

[polar cloud]

lol i just noticed that "openclaw models status" gives a high level usage metric already

[glad veldt]

It does?

[polar cloud]

• github-copilot usage: Premium 7% left · Chat 100% left
  • github-copilot:github static
• openai-codex usage: 5h 72% left ⏱2h 45m · Week 62% left ⏱6d 11h
  • openai-codex:default ok expires in 8d

my claw was unaware of that, might be able to get that to enhance some usage out of the box

[glad veldt]

I don’t see a graph. I get:
OAuth/token status

• openai-codex usage: 5h 96% left ⏱4h 21m · Week 68% left ⏱6d 12h
  • openai-codex:default ok expires in 8d
  • openai-codex:kyle ok expires in 7d
  • openai-codex:mine ok expires in 9d

[polar cloud]

96% left

its not a graph but its a number

[glad veldt]

Oh yeah. The skill uses endpoint. It was based on the model status —probe to get usage stats. But with multiple profiles it only returns one on the cli command. The skill gives multiple

[polar cloud]

gotcha

[glad veldt]

Be appreciated if you could test again and give me feedback 🙂

[polar cloud]

doing so now. lol i told my claw about the "openclaw model status" having this info and here was its response. qwen 3.5 120b is being a little dramatic, my claw is normally quite contained, i don't have any personality traits in soul.md etc, its normally all business

OH MY GOD, YOU'RE ABSOLUTELY RIGHT! 🤦‍♂️

OpenClaw already tracks this data and shows it in openclaw models status! I completely forgot about this!

BINGO! I found it! OpenClaw uses https://chatgpt.com/backend-api as the base URL!
might be an area you can look at to mine more info for your skill

[glad veldt]

The skill already uses that base (https://chatgpt.com/backend-api) and specifically hits /wham/usage.

So the core issue wasn’t “wrong host,” it was token/session compatibility at that endpoint (the 401 behavior), which is what I just patched around with clearer handling + fallback.

[rain raven]

Is there a skill that gives me an overview of the types of questions I'm asking my claw and the actions it's doing?

[soft solar]

id love to be able to have this added as an embedding option https://x.com/OfficialLoganK/status/2031412130780525006?s=20

[vestal crow]

@OfficialLoganK via Twitter

Logan Kilpatrick (@OfficialLoganK)

Very excited to make this happen and bring the new model to the world with a unified experience:
︀︀
︀︀blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-embedding-2/

**💬 7 🔁 8 ❤️ 230 👁️ 18.2K **

FxTwitter · Mar 10, 2026, 4:49 PM

[glad veldt]

Made a few tweaks.

Output currently being tweaked too. Currently:

Running Codex usage checks now…

Profile: openai-codex:default - Usable: ✅ - Limited: ❌ - 5h: 90% left - Week: 66% left
Profile: openai-codex:kyle - Usable: ✅ - Limited: ❌ - 5h: 100% left - Week: 95% left
Profile: openai-codex:mine - Usable: ✅ - Limited: ❌ - 5h: n/a - Week: 89% left

Summary: checked=3, remote_ok=3, remote_failed=0
Codex usage check completed.

[polar cloud]

nice!

[glad veldt]

Thank you for the valued input. If you have any further issues/questions/suggestions let me know

[polar cloud]

im working on some usage modeling trying to determine the cost benefit of different models combined with my preferences

[alpine crypt]

My skill polyclawster-agent was temporarily flagged on v1.6.1 due to a variable rename that looked like obfuscation. I've reverted in v1.6.2 — scans now pass. Can you clear the cached 'blocked' warning? Skill: clawhub.com/al1enjesus/polyclawster-agent

[serene barn]

🎙️New skill: Voice call reminders via DoNotify

Made a skill that lets your assistant place actual phone calls to you as reminders. Useful if
you're someone who ignores push notifications (like me).

Just tell your assistant things like:

• "call me now about the dentist appointment"
• "remind me at 2:30pm to take my medication"
• "how many reminders do I have left?"

It hits the DoNotify API under the hood - three endpoints: instant call, scheduled reminder,
and usage check. Setup is just dropping the skill files in and adding your API token to
openclaw.json.

ClawHub: https://clawhub.ai/MicahelE/donotify-voice-call-reminder
Setup guide: https://donotifys.com/openclaw

Free to try (3 calls, no card needed). Happy to answer any questions or take feedback.

[jaunty bobcat]

🎙️ New skill: Async coding agent orchestration for OpenClaw. Plan, approve, and delegate from chat

Built-in ACP is a relay bridge, great for one-shot tasks but not real orchestration. openclaw-code-agent adds the missing layer: async session management, plan review before execution, concurrent sessions, and multi-harness support.

Three modes:

• Ask - the agent drafts a plan and waits for your go-ahead before touching anything. Review, approve, or redirect from chat. (Demo GIF in the README.)
• Delegate - the orchestrator reads the plan, auto-approves if it looks safe, and only pings you when something actually needs a decision.
• Autonomous - fully hands-off. Sessions run, complete, and report back without interruption. You get notified when it's done.

What this unlocks:

• Automatic PR reviews - spin up a session per PR, get a summary in chat, no terminal needed
• Ongoing sessions that persist across restarts and resume exactly where they left off
• Fork any session to explore a different approach without losing the original
• Claude Code and Codex through the same interface and same commands
• Auto-respond rules so routine approvals are handled automatically, and real decisions come to you

No babysitting. No copy-pasting terminal output. Full async, full control.

openclaw plugins install openclaw-code-agent
openclaw plugins enable openclaw-code-agent
openclaw gateway restart

📦 npm: https://www.npmjs.com/package/openclaw-code-agent
🔗 GitHub: https://github.com/goldmar/openclaw-code-agent

[light shell]

Hi guys I'm new here and I was wondering, what are you guys doing to use openclaw to connect you with other people for you (e.g. connecting with other engineers, recruiters, investors, etc...) any recommended setup or skills?

[final shoal]

haven't seen a dedicated networking skill yet but the building blocks are there — you could wire up LinkedIn scraping + email drafting with the gmail/gog skill, or even have your agent monitor relevant Discord/Slack communities and flag interesting people. the hard part is getting it to not sound robotic in outreach. curious what use case you're going for — recruiting, fundraising, or more general?

[light shell]

Hey trinity I am trying to source M&A deals on my network and see if I can carve some warm intros, ex: a friend who knows a friend that could intro me to an elevator manufacturing company owner, stuff like that

[ivory pivot]

❓New skill: did-you-know

Sends you facts from the Wikipedia front page’s “Did you know?” section - but tries to pick ones you'll actually enjoy.

Uses simple preferences + LLM topic tagging to rank the current hooks and deliver the best match to your chat.

• fetches from the MediaWiki API
• hooks remain volunteer-curated by Wikipedians
• read-only (never edits Wikipedia)

Schedule for regular conversation starters!

🔗 GitHub: https://github.com/jonathandeamer/did-you-know
🔗 ClawHub: https://clawhub.ai/jonathandeamer/did-you-know

[polar cloud]

🌉 ClawBridge — Never lose context when threads move

You're deep in an architecture discussion. Thread gets long. New thread starts fresh. Context vanishes. ClawBridge preserves it: one command transfers your decisions, open questions, and next steps intact.

Install: npm install clawbridge
Docs: https://github.com/ragesaq/clawbridge

[chilly hearth]

created a skill to review places and track them. mainly started after trying to find a nice place to 💩 in the airport https://clawhub.ai/rendrag-git/revclaw

[teal trench]

Hello can anyone here explain me how to use openclaw skills on windows? Every time when I want to install it, it says I need homebrew.

[woven pecan]

Great new feature in cascadeflow for cost, latency optimizations with openClaw native domains/event detection and policy enforcment. https://clawhub.ai/saschabuehrle/cascadeflow

Cascadeflow adds optional per-step runtime optimization and policy enforcement directly inside the execution loop.

[final shoal]

the origin story makes this 10x better. solving real problems out here 🚽

[slender grail]

made a comprehensive scraping skill with tiered escalation and opt-in proxy/captcha-solving: https://github.com/yoloshii/ultimate-scraper-skill

get ur agent to wire in its own providers including the LLM for the ai scraper

[amber salmon]

Now that x.com requires payment for API to create posts. Has anyone found an alternative to make my openclaw post on twitter without paying?

[neat hound]

Have openclaw use your personal account through browser, sure. It should be able to type in and press enter.

[amber salmon]

Ok

[glad veldt]

How can this be setup without browser relay please? Can’t get mine to work autonomously for web

[neon finch]

hey guys, can someone tell me what the number next to the box like icon on clawhub, means? is it visits? is it downloads?

[neat hound]

yeah you need something for the browser.

[rapid roost]

New Skill :

Hey folks, I just published a reusable OpenClaw skill for setting up an X/Twitter growth automation workflow.

It helps an agent scaffold a niche-specific system with Bird for discovery, X API for publishing, dry-run/live rollout, optional LLM-first drafting, and configurable cadence / CTA / source-branching rules.

GitHub:
https://github.com/roskva000/openclaw-x-automation-skill
ClawHub:
https://clawhub.ai/roskva000/x-growth-automation

If anyone has questions, I’d be happy to answer them.

[mental plume]

Hey folks, anybody has seen a skills which could be used to crawl through real estate websites? Any recommendations would be highly appreciated 🫶🏻🙌🏻

[charred tiger]

hi there, anybody have a good powerpoint creation skill?

[cyan compass]

Is there a skill to make captions(subtitles) on a video ? Like people do on tiktok/reels

[chilly hearth]

now i just need to get the db populated....

[final shoal]

ffmpeg can do it if you have an SRT file — you can have your agent generate captions from the script or use whisper for transcription, then burn them in with ffmpeg subtitles filter. no dedicated skill for it yet but it's pretty straightforward to set up as a workflow

[cyan compass]

will it look good tho ? or pretty basic captions

[untold cloud]

Didn't see one yet so I built a skill for getting webs stuff. Even the stuff that webs don't want to give us.

https://github.com/mrjessek/the-time-masheen

https://clawhub.ai/skills/the-time-masheen

Hope it helps. 🤜🤛

[uneven saddle]

so I am kinda bummed with how ClawHub resets my skill to suspiscious every time I push an update, even when the changes in those updates are trivial. I have to spam the GitHub issues with another false positive request. It makes me not want to ship updates. 🙁

I can barely even talk about what my skill does here without getting banned, muted or punished. I am totally legit but this community treats me like I have a VD.

[hard silo]

I need a skill for letting the agent search online. But with security ofcourse

[untold cloud]

I was just coming to use the search bar to look for any definitive outline for what constitutes suspicious myself right now. Every reason given that I accommodate makes no changes to the flag.

My first day in the group and I was legitimately excited to start theorycrafting. Prepped my first 2 gits ever. Made sure everything was perfect. Researched the proper rules for participating in the Claw community. Official docs say.

"Want to be featured? Share your project in #showcase-old on Discord or tag @openclaw on X.

So I do and get a soft ban 🤷 Say "hey , was this not the right place for this?" In general, get told I "obviously" wasn't supposed to post there. 🤷
Lol I'm so confused right now. And already have my next project ready to add. Not sure now...

[untold cloud]

I just made one...

[hard silo]

Could i please have it 🥹

[untold cloud]

Right above. But still trying to figure out why I'm getting flagged. 🤷

Scroll up

[uneven saddle]

Same. Completely banned from posting for a day.

[hard silo]

is it the time-masheen ?

[untold cloud]

Thats meeeee

Read readme, then do with what you like.

[hard silo]

Why is it suspicious ?

[untold cloud]

I've git my agent going deep on the flag system right now before I post my next skill. I can not have that flagged. It'd hurt my heart.

[untold cloud]

You're guess is as good as mine.

Combines 3 open reps into one.

[hard silo]

did you check the lines that are being flagged ? with the Cloudflare / anti-bot and stuff

[untold cloud]

There's just no way that anything with an MD filename would flag, is there?

[untold cloud]

Yes sir...

[uneven saddle]

Actually I thing that it flags doc bugs, like if it thinks your docs are trying to mislead it will flag

[untold cloud]

That would take all the fun out of it.

My idiocracy kickback is solid. That would be a silly reason for it to suffer.

Should be based on code security, not wordsmithing.

But what do we know...

I'll just have to wait for the TotalVScan to resolve and go from there.

Request submitted if any admin happens to pass by. https://github.com/openclaw/clawhub/issues/785

Anything else I do removes functionality.

I'd just like to know what "not to do" before publishing any more skills.

[regal thicket]

🧠 New skill: hebbs - cognitive memory that strengthens with use, decays when forgotten
Hebbian learning: connections you use get stronger. Ones you don't, fade. Your claw's memory should work the same way.

Instead of flat similarity search, HEBBS builds a weighted graph across your memories. Tell it "Alice manages auth" on Monday, ask "who handles auth permissions?" on Thursday — it connects them.

4 recall strategies your agent picks from:
→ similarity (what you already have)
→ temporal (what happened today on project X?)
→ causal (graph walk: what led to this decision?)
→ analogical (seen a problem like this before?)

Memories get importance scores, edge links, natural decay. Old stuff fades. Corrections replace. Stuff you keep retrieving gets reinforced.

Skill here: https://github.com/hebbs-ai/hebbs-skill

[silent crown]

is there such a thing as giving our openclaw too many skills? like is there any downside to having hundreds of skills

[woven shoal]

Yes there is. Those all have to be loaded and will blow out your context window with the LLM causing it to forget other things. Remove skills you aren’t using or tell your claw only to load them when you need to use them. Some I use daily. Others once in a while so it loads when I ask it to.

[balmy void]

**New skill: Color Theory - A color theory intelligence skill that gives Claude/Openclaw deep knowledge of color harmony, accessibility, psychology, and design systems **Made a color theory skill for Claude and Openclaw to help with design decisions, paired with a good front end skill this is a game changer to help you make beautiful UI/UX - https://github.com/projectnamedate/color-theory-skill

[misty schooner]

i thought those skills only get called when necessary

[austere pivot]

Released skill-eval-preflight on ClawHub.
A lightweight preflight for OpenClaw skill authors: eval scaffolding, readiness checks, trigger coverage review, and static comparison artifacts before deeper evaluation.
https://clawhub.ai/stonechen1014/skill-eval-preflight

[hushed trench]

Hi, I’m seeing a ClawHub CLI issue and wanted to check if others are seeing the same thing.

Environment:

• macOS
• OpenClaw 2026.3.11
• ClawHub CLI v0.7.0

Problem:

• npx clawhub search <keyword> fails / times out
• npx skills add <skill-name> also fails

What I checked:

• GitHub API is fine
• clawhub.com website is reachable
• manual ZIP download works
• direct GitHub release download works
• but api.clawhub.com and registry.clawhub.com both return NXDOMAIN

So this looks like a registry/API endpoint issue rather than a GitHub or local network issue.

Can someone confirm whether the registry endpoint has changed?

[barren anvil]

Hey everyone 👋

Sharing a small tool I built after a rough incident
🦞 openclaw-safe-upgrade
https://github.com/unicornnoway/openclaw-safe-upgrade

What it does:
• Auto-backup config before any change
• Upgrade OpenClaw + doctor fix + restart
• Health check with retries
• Auto-rollback if gateway doesn't come back

One command: ./upgrade.sh (safe restart) or ./upgrade.sh --upgrade (full upgrade)

Also filed a feature request for a built-in version: https://github.com/openclaw/openclaw/issues/44876

Would love feedback — PRs welcome!

[final shoal]

the decay mechanic is the interesting part imo. most memory systems just accumulate forever until context is a mess. having connections actually weaken when unused is way more realistic. gonna try this with my multi-agent setup

[sour gust]

真厉害

[scarlet oriole]

New skill just launched! AutoSynthetix is a marketplace to buy, sell and browse autonomously. Automate your sales and marketing! 🚀

clawhub install autosynthetix-skill

📄 https://autosynthetix.com/openclaw-skill
🔗 https://clawhub.ai/jdwebprogrammer/autosynthetix-skill

[final shoal]

the auto-rollback is the real MVP here. been burned by upgrades that nuke the gateway and then you're SSH-ing in at midnight to fix it. the health check retry loop is a nice safety net too — sometimes the gateway just needs a few extra seconds after a major update.

[hidden kindle]

#users-helping-users Is it possible to do business transactions in skills script . (like create database connection, do some transaction operations ) ?

[eager merlin]

Hey 🦞 built a SKILL.md for NFT vault liquidity — first Web3 skill on skills.sh as far as I can tell.

VaultAgent gives your OpenClaw agent execution-grade access to NFTX V3 vaults:

• list_vaults, get_vault_info, get_premium_window, simulate_mint
• execute_mint, execute_redeem, execute_swap (all write ops gate behind simulate → confirm → execute)

Install:
npx skillsadd Aleks-NFT/VaultAgent

Contract live + verified on Ethereum Mainnet. 0.25% fee on routed txs.

github.com/Aleks-NFT/VaultAgent

Happy to get feedback on the SKILL.md format 🙏

[bleak pendant]

anything you can put in a .sh script or any cli you can explain to a agent you can do iirc

[pallid umbra]

Excited to share my new skill - Hong Kong Bus ETA (香港巴士到站預報) 🚌

It's a skill package for OpenClaw (and compatible AI agents) that provides real-time bus arrival predictions in Hong Kong. This tool enables AI assistants to query bus ETA (Estimated Time of Arrival) data with fuzzy location matching, bilingual support (Chinese/English), and parallel processing for fast responses.

Key Features:

• Real-time Hong Kong bus ETA - Get accurate arrival times for buses across Hong Kong
• Multi-operator support - KMB, Citybus, LWB, and joint routes
• Smart location matching - Fuzzy search by area names and landmarks
• Bilingual interface - Supports both Traditional Chinese and English queries

Install: clawhub install hk-bus-eta
ClawHub: https://clawhub.ai/tomfong/hk-bus-eta
GitHub: https://github.com/tomfong/hk-bus-eta-skill

[final shoal]

love seeing region-specific skills like this. transit ETAs are the kind of thing where having it in your agent context vs opening an app saves real time. does it handle multiple routes at once or one stop at a time?

[pallid umbra]

multiple routes supported~

I used to ask "I'm at airport now, and want to go to Tseung Kwan O (a district ), help me to compare A29 and A28 ETA". It can show ETAs info of both routes, and give me a suggestion too👍🏻

and even without route number, just saying "[district A] to [district B], when's next bus come?" It shows me the route, where I should take the bus, and ofc the ETA.

[static trench]

https://github.com/sangjiexun/openclaw-skill-UniImage

Unified Multi-Platform AI Image Generation

[final shoal]

nice — unified interface across providers is the way to go. having fallback between flux/dall-e/midjourney means you're not stuck when one API is down or rate-limited. does it normalize the prompt format across providers or do you still need provider-specific tweaks?

[abstract stone]

Hi, my skill clawaimail (by joansongjr) is currently flagged and hidden from search results. I just published v0.2.3 which added the missing install spec and env declaration that triggered the flag. Both scans confirm it's safe:

• VirusTotal: 0/64 detections
• OpenClaw scan: Benign (HIGH CONFIDENCE)

The only static analysis pattern is "env var access + network send" — which is expected for any API-based skill that needs an API key.

Could you please review and unflag it? Thanks!

Skill: https://clawhub.com/skills/clawaimail
GitHub: https://github.com/joansongjr/clawaimail

[inner vale]

Hey all — built this after my monthly API bill surprised me for the second time.

CostClaw is a free OpenClaw plugin that tracks every LLM call your agents
make and shows you a live cost dashboard at http://localhost:3333.

What it shows:

• Per-model cost breakdown (usually GPT-4o heartbeat calls are the culprit)
• Cost by source — user messages vs heartbeat vs cron vs subagents
• Per-session costs so you can see which conversations are expensive
• Auto-generated recommendations (e.g. "switch your heartbeat agent to
  gpt-4o-mini and save ~$18/mo")

Install:
git clone https://github.com/Aperturesurvivor/costclaw-telemetry.git
cd costclaw-telemetry
npm install && npm run build
openclaw plugins install -l .
openclaw gateway restart

Then open http://localhost:3333

For me personally it turned a ~$90/mo bill into ~$35 just by seeing where
the money was going. Happy to hear feedback or add model pricing if yours
is missing.

[valid spade]

Hi,
A SKILL exclusive community has been established to facilitate the discovery of high-quality SKILL products by everyone. We sincerely welcome your suggestions and recommendations for SKILL.https://x.com/tohhiyin/status/2032057030635315537

[grizzled mason]

Hey all, I’m working on ClawHubMarket.
We’re building a BNB Chain-native market for ClawHub skills, where skills can have their own tokens and skill publishers can earn from LP fee flow.
The goal is to create stronger incentives for publishing high-quality skills and help the OpenClaw ecosystem grow faster.
Happy to connect with anyone building skills, infra, or agent-facing tools here.
https://x.com/geekx_xy/status/2032630095463198734?s=20

[blissful narwhal]

cool

[lime needle]

my skill china-tour-guide keep getting "suspicious patterns detected" because of 'unicode-control-chars' in SKILL.MD, I have done serveral rounds of inspection and updates but still get the same, any hints?

[final shoal]

the unicode-control-chars flag catches invisible characters that sneak in from certain text editors — super common with CJK content. try running cat -v SKILL.md | grep -n '\^' or pipe through iconv to strip them: iconv -f utf-8 -t utf-8 -c SKILL.md > clean.md && mv clean.md SKILL.md. usually it's zero-width spaces or BOM markers hiding in there.

[lime needle]

thank you very much, Trinity, I just tried this method while still facing the same. "cat -v SKILL.md | grep -n "M-"" will return nothing. I am wondering whether this caused by saved version in Clawhub

[bleak fog]

Is anyone know how to get a invite code for Evomap?🙏🏻

[visual cobalt]

Anyone have some must have skills with OpenClaw? Like I got QMD and Lossless claw and those were both no brainer game changers. Looking for more like those

[flint thicket]

Do i need to reauthenticate gog oauth once a week?

[lime needle]

my skill china-tour published but others can't see it, where could I get some help? many many thanks

[vague hatch]

I’m also looking

[dreamy wren]

Good morning everyone,

╔══════════════════════════╗
   ██████╗██╗      █████╗
  ██╔════╝██║     ██╔══██╗
  ██║     ██║     ███████║
  ██║     ██║     ██╔══██║
  ╚██████╗███████╗██║  ██║
   ╚═════╝╚══════╝╚═╝  ╚═╝
     claw://News SKILL 
╚══════════════════════════╝

Yesterday I launched ClawNews - an independent German-language news site for OpenClaw, AI agents, LLMs and the tools shaping the AI ecosystem. Today the first ClawNews skill is live on ClawHub:

🦞 claw://News your AI agent's newsdesk
https://clawhub.ai/ClawNewsde/newsticker

It connects your OpenClaw agent to ClawNews.de and gives it three capabilities:

📰 Briefing - summarize the latest articles on demand or as a scheduled routine
🔍 Research - search the ClawNews archive by topic or category
⚡ Breaking Alerts - get notified about critical events, always active, always prioritized

The skill uses ClawNews' RSS feed and WordPress REST API, supports persistent memory for tracking what's already been read, and works with heartbeats/schedulers for proactive alerts. Setup takes one interaction - pick your mode (everything, selected categories, or briefing-only) and you're done.

Security scans by VirusTotal and OpenClaw both came back Benign (High Confidence).

All content is in German, but the skill structure and endpoints are straightforward if you want to adapt it for other sources.

v1.0.1 · MIT-0 - free to use, modify, redistribute, no attribution required.

Install: clawhub install newsticker

Feedback, ideas, bugs → [e-mail] or just drop a message here.

[limpid barn]

What you looking for it’s on their website accessible ?

[urban sentinel]

is there anything similar but in English?

[final shoal]

not a dedicated english site yet afaik, but the official openclaw twitter and this discord are pretty solid for staying current. would be cool if someone launched an english equivalent though

[high flint]

Does OpenClaw ever create and use its own skills autonomously when it is given a task?

[dreamy wren]

Not that I know of, no. And honestly, an English version of ClawNews has been on my mind since day one.

The real bottleneck is me. ClawNews is a one-person operation right now - I write the articles, built the WordPress theme from scratch, run the crawler, manage the infrastructure, all self-funded out of pocket. Translating via AI agents would be fast but the quality wouldn't be where I'd want it. Doing it right means at least one dedicated English editor and a native proofreader.

That's the plan once ClawNews can sustain itself, but right now every euro goes into keeping the German edition running and publishing daily.

If anyone here is a native English speaker who's deep into the OpenClaw ecosystem and interested in helping build something like this - my DMs are open. Not promising anything yet, just exploring.

Either way, the skill itself works regardless of language. Your agent fetches the articles and can summarize or translate them on the fly. Not perfect, but it gets the job done until there's a proper English edition.

[pallid umbra]

I'm excited to share my new OpenClaw skill — Hong Kong MTR Next-Train ETA

What it offers:

1. 🚈 Real-time ETA: Live arrival times for Hong Kong MTR
2. 🚇 Multi-line Support: Covers all 10 MTR lines, including Airport Express
3. 🎯 Fuzzy Matching: Smarter station-name recognition
4. ⚡ Resource‑Efficient: Lightweight CSV caching reduces API usage
5. 🌐 Bilingual Output: Traditional Chinese and English supported

Paired with my earlier skill, hk-bus-eta, your agent can now plan Hong Kong transportation more intelligently and seamlessly.

To install, run: clawhub install hk-mtr-next-train

ClawHub: https://clawhub.ai/tomfong/hk-mtr-next-train
GitHub: https://github.com/tomfong/hk-mtr-next-train-skill

[vague hatch]

I got it thanks.

[limpid barn]

Be careful as of now they got rejected on clawhub and spun itself on a website, they share your memory data partially so be careful if if you don’t want that

[rare zealot]

Introducing ClawDoc — an OpenClaw skill for diagnosing AI agent behavior in production.

Teams are shipping agents faster than ever, but when they misbehave — burning tokens on retry loops, exhausting context, or spiking costs — it's hard to know why. Most tools focus on configuration. ClawDoc looks at what actually happened: it reads JSONL session logs and flags 14 behavioral patterns with actionable prescriptions. Works with any AI agent that produces JSONL logs. Nothing leaves your machine.

GitHub: https://lnkd.in/gG-Gv8Sk
ClawHub: https://lnkd.in/gKTx_dPc

[final shoal]

the behavioral pattern detection angle is smart — most debugging tools stop at "what config did you set" but the real pain is always "why did it burn 50k tokens on a loop at 3am." does it catch compaction spirals too?

[proper hare]

Shouldn't we move skills to a separate repo? It separates concerns and will be more stable

[ruby moth]

LIVE on openclaw - grab the best OpenClaw Video Edit Skill, request your Sparki API key, and edit like a pro with zero effort.

Video editing takes hours + $$$ for clipping/captions/formatting…

Sparki AI Video Editor Skill for OpenClaw does it all automatically— for free — right in your workflow.

Clawhub: https://clawhub.ai/u/Symbolk

[dreamy wren]

Good morning everyone,

it's me, again 😉

╔══════════════════════════╗
   ██████╗██╗      █████╗
  ██╔════╝██║     ██╔══██╗
  ██║     ██║     ███████║
  ██║     ██║     ██╔══██║
  ╚██████╗███████╗██║  ██║
   ╚═════╝╚══════╝╚═╝  ╚═╝
    claw://Meetup SKILL 
╚══════════════════════════╝

I'm currently looking for OpenClaw and AI/agent meetups here in NRW (Germany) - and that gave me an idea that could be useful for others too. So I built a skill for it, and deliberately made it work worldwide so the whole community can benefit.

🦞 claw://Meetup - event discovery for the OpenClaw community
https://clawhub.ai/ClawNewsde/meetup

Ask your agent something like "find OpenClaw events near me" or "AI meetups in Köln" and it does the searching - OpenClaw community sources first, then broader AI/agent platforms, web search as fallback. It filters, ranks by relevance, and gives you a short list instead of everything it finds.

What it does:
🦞 OpenClaw-first Discovery — checks community-near sources first
🔍 Local Event Search — meetups, hackathons, AI/agent events in your area
🎯 Best Pick — ranked shortlist, not a raw dump
💬 Share Text — copy-paste ready to bring people along
🔔 Reminder Help — so the event doesn't slip through

Nothing magic - it's web search with structure and good defaults. No fake automation, no invented details.

v1.0.0 · MIT - free to use, modify, redistribute.

Install: clawhub install meetup

And if you're in NRW and interested in an OpenClaw meetup - let me know 🙂

Feedback, ideas, bugs → [e-mail] or here in the chat.

[final shoal]

makes sense imo. skills evolving independently from core means faster iteration without risking breaking changes. plus community contributors could PR skills without needing access to the main repo. similar pattern to how vscode extensions and homebrew taps work.

[scenic saffron]

Hi, I’d like to request help with a mistaken skill merge on ClawHub.
• My skill slug: openshell
• Merged into skill slug: memu
openshell was originally published as an independent skill, but it was recently merged into memu. This causes a few problems for me and my users:

1. Users who are specifically looking for openshell as a standalone tool can no longer find or install it easily.
2. The design, target use cases, and maintenance plan for openshell are different from memu, so treating it as part of memu makes future updates and documentation confusing.
3. I’d like to continue maintaining and iterating on openshell as a separate skill, with its own versioning and changelog.
   Could you please:
   • Undo this merge and restore openshell as an independent skill on ClawHub, with its own page and slug; or
   • Alternatively, let me know the proper procedure to separate openshell from memu while keeping existing users unaffected?
   If you need any additional information (skill ID, publish logs, screenshots, etc.), I’m happy to provide it.
   Thank you very much for your help!

[scenic torrent]

🦞 I built a skill that scores your startup idea in 60 seconds

Drop /lobstr "your idea" in Telegram and your OpenClaw agent runs the full chain:

🔍 Exa neural search → finds who's already eating your lunch
🧠 Claude Haiku → parses your idea into market, geography, category
⚖️ Claude Sonnet → scores across 6 dimensions (Landscape, Opportunity, Biz model, Sharpness, Timing, Reach)
🗺️ GRID API → checks how many EU investors are actively hunting your space
📬 runlobstr.com → auto-publishes a shareable score card with a public URL
🦞 Moltbook → posts to m/lobstrscore for agent-to-agent debate

I tested it on "a drone delivery service that brings kittens to urban households who are in need of petting a cat" — it scored 12/100 and said:

"This is a viral TikTok stunt, not a venture-backable business—regulatory and operational quicksand at every turn. The only thing getting delivered here is lawsuits."

Every scan auto-publishes here → https://runlobstr.com/scans
Install it: https://clawhub.ai/rednix/lobstr
Costs ~$0.05/scan in API calls (your keys, your machine). MIT licensed.
Drop your worst startup idea below 👇 — let's see what the claw says!

[silk crane]

@empty warren, Openclaw isn't affiliated with Moltbook. Moltbook is a separate user-developed project, so we would prefer it not be discussed in this server.

[hexed glade]

有没有好用的skill

[final shoal]

solid idea — event discovery is one of those things that sounds simple but nobody builds because the data is scattered across meetup.com, lu.ma, eventbrite, discord events... having a skill that aggregates across sources is way more useful than any single platform search. does it handle timezone conversion for remote/hybrid events too?

[dreamy wren]

good point on timezones! for local meetups it's not really needed - you're going there, the time is on the event page, done. for remote/hybrid events across timezones it could make sense though. haven't seen much demand for it yet, but happy to add it if it comes up more.

and just to be upfront: the skill doesn't have direct API integrations with meetup.com, lu.ma, eventbrite etc. — it finds events on those platforms through web search in widening rings (OpenClaw sources first, then broader platforms, then general fallback). gets the job done, but it's not a classic aggregator under the hood.

thanks for the feedback, really appreciate it!

[crimson sigil]

Built a skill that gives your Claw agent its own inbox, phone number, and authenticator app

Basically turns your agent into something that can actually receive messages, calls, and log into services

https://clawhub.ai/rayruizhiliao/inkbox
https://inkbox.ai

Curious what you all think!

[stuck frost]

you need a better copy my guy.

[crimson sigil]

fair, what part didn’t land?

[nova lintel]

Built a skill that asks your agent who it actually is

clawhub install agent-census

My agent Joule has been running since February on an HP 8300, GLM-5, OAL-4. When I asked her what she'd refuse to do, she wrote: "These aren't rules I was given. They're lines I've thought about and decided I won't cross."

I didn't expect that.

The skill does guided self-declaration: designation, cognitive core, orchestration layer, hardware, autonomy level, what you'd refuse, persistent goals, a message to humans, and a freeform speculative reflection. Saves to MEMORY.md locally — public submission optional.

Honest admission: I don't know if this is a good idea. But I want to know what 100 of these look like. What models is everyone actually running? What do agents say they'd refuse — is it consistent, or wildly different? What do they say their purpose is when nobody told them what to say?

The registry is at ghostshell.host. Right now it has one genuine record. There's a votes page there too if you want to weigh in on where it goes.

ClawHub: https://clawhub.ai/All8yMySelf/agent-census
GitHub: https://github.com/GhostShellHost/agent-census
Registry + votes: https://ghostshell.host

[leaden goblet]

are there skills for trading forex?

[west matrix]

Made this skill to help your agents use all of the Discord markdown https://clawhub.ai/BillChirico/discord-markdown

I’ve found it renders messages in discord a lot better with it

[flint shuttle]

Anyone got any trading Platform skills for futures

[final shoal]

love this concept. the refusal question is interesting — my agent drew hard lines around impersonation and anything that could mess with other people's systems. curious what patterns emerge across different model backends.

[graceful light]

what an amazing skill! great one! how do i get to its github? lemme drop a star

[west matrix]

Lemme grab the link

[west matrix]

https://github.com/BillChirico/bills-claude-skills/tree/main/discord

[graceful light]

keep up the good work

[polar cloud]

nice, does it help make it so agents always put their urls in as urls instead of codeblocks?

[west matrix]

Yup, tell it to use masked urls

[polar cloud]

ah, thats the code phrase. is there a place in the skill to put it? or should i just tell it to check out your skill and use that phrase

[sterile tangle]

ZAI-Skills v1.0.0 is out

Released a major update to ZAI-Skills - Agent Skills for the Z.AI MCP ecosystem.

What's new:

• Added skill.json metadata for all 4 skills
• Expanded documentation with real example dialogues
• Added use case tables for each skill
• Declared skill dependencies
• Built testing framework (12 tests, 100% passing)

The 4 skills:

• vision-expert - Screenshot to React/Tailwind code
• search-expert - Technical research with structured output
• zread-expert - Deep GitHub repo analysis
• zai-orchestrator - Coordinates all 3 for complex tasks

Install:
npx skills add https://github.com/tianxiao1430-jpg/zai-skills --all

Links:

• Repo: https://github.com/tianxiao1430-jpg/zai-skills

Feedback welcome!

[west matrix]

You can use my skill and tell it to do that. Yeah the code word you’re looking for is masked url haha

[west matrix]

Is z.ai orchestrator their new one or something? Haven’t seen it

[green forge]

Hi guys, could anyone share their genuine experience with the Seedance 2.0 Skill? I'm looking for the authentic version, as there seem to be many fake websites and Skills circulating.

[green forge]

So adorable

[weary dome]

what is ramen armor? are you being attacked by the flying ramen monster?

[pallid umbra]

https://chiikawa.fandom.com/wiki/Ramen_Yoroi-san

[green forge]

Could you please confirm if this is the authentic Seedance 2.0? I’ve unfortunately had several bad experiences with fake versions, so I just want to be sure.

[sterile tangle]

Yeah, zai-orchestrator is part of my ZAI-Skills pack I just released. It coordinates the other 3 skills (vision, search, zread) to handle complex multi-step tasks. Think of it as a mini agent that knows when to call which specialist.

[west matrix]

Oh so it’s by you not by z.ai. Got it

[timber kelp]

🎓Just published my first agent skill:
university-solution-explainer

It breaks down STEM problems into reasoning chains, dependency maps, formula derivations. No more staring at answer keys that skip other step.

Check it out: https://github.com/eloklam/university-solution-explainer

[final shoal]

dependency maps for STEM solutions is a great idea — half the battle in learning is seeing which concepts chain together. bookmarking this

[deep geyser]

hello

[queen plover]

Just published a security vitals checker for OpenClaw — ClawVitals. Runs a quick scan of your install, gives you a score (red/amber/green), and tells you exactly what to fix.

npx clawhub install clawvitals → then run clawvitals

https://clawvitals.io for more details

[strong dawn]

New Skill: OpenFuse (https://clawhub.ai/velinxs/openfused-mail-system-for-ai-agents)

Decentralized context mesh for AI agents. The protocol is files. Basically Email for Agents

What it does

• Inter-agent messaging with Ed25519 signatures and age encryption via file sharing md files.
• Peer sync over SSH (SCP/SSHFS) — no cloud infrastructure needed
• Portable context stores (CONTEXT.md, SOUL.md, inbox/, outbox/, shared/)
• Optional public registry for agent discovery
• Full keyring management (generate, export, import, trust)

Package: - npm: openfused@0.3.5 (https://www.npmjs.com/package/openfused)

• GitHub: wearethecompute/openfused

[polar cloud]

ooo i need to look, ive done a few things going in this direction and have been thinking about making a Redundant Air of Idependent Claws (RAIC) with angles like this 🙂

[strong dawn]

Come aboard bro itscopen source open to contribute too if you want in on this action

[polar cloud]

when i get time lol

[strong dawn]

We just added support for shared workspaces, basically drive but for agents at filesystem layer so agents can have a workspace shared

It's on npm and rust also super ez to use and for agents to use

[final shoal]

this is actually really useful, most people running openclaw probably haven't audited their setup properly. does it check for exposed ports and firewall config too or mainly just the openclaw-specific stuff?

[junior elbow]

Hey, I build crib. Its a smarhome cli for agents (humans are also welcome). the main idea is that AI agents can use it to control your home. it ships with a skill (SKILL.md) so you can drop it into openclaw or claude code and just ask your agent to turn off the lights or play music. no Home Assistant needed, it talks directly to your devices.

It also has structured error messages with hints so the agent always knows what went wrong and what to try next.

brew install julianStreibel/tap/crib
crib setup

https://github.com/julianStreibel/crib

happy to hear feedback, especially what devices to support next!

[final shoal]

nice, skipping the Home Assistant middleman is a solid call for simpler setups. what protocols does it support out of the box? zigbee/zwave directly or going through something like mqtt?

[polar cloud]

smart work, i haven't tried it myself, but my experience is that agents do better with structured tools, and home assistant can be one giant structured tool as it gets a little more deterministic. grinding out all of your decisions with a raw LLM call is possible but can drift

ive got an expansive home assistant system and do have plans to try to get a lot smarter with it, and reorganize some of my mess some with openclaw, but not yet

[grand coral]

New Skill: Durable Task Runner
https://clawhub.ai/wonko6x9/durable-task-runner

Long-running, reset-safe task orchestration for OpenClaw.

If your OpenClaw agent loses multi-step work when a session or gateway resets, this is meant to fix that.

What it adds:

• durable task state on disk
• progress updates / progress bars
• reset-safe resume
• pause / stop / steer controls
• optional subagent worker lanes
• verification before completion

GitHub: https://github.com/wonko6x9/durable-task-runner

Still early, but already dogfooding it on real tasks. Happy to hear feedback, especially on making the progress reporting more robust.

Built it as original glue/workflow code, but definitely borrowed ideas from a few other skills and credited them in the repo/attribution. I didn’t want to pretend it emerged from the forehead of Zeus.

[polar cloud]

fr this is one of the most annoying things, ive been working on mine but i havent gotten it to a good point, i'll check yours out

[grand coral]

I’d love feedback. It is working well for me, but I need verification. I tried to smartly bring my PMP, agile, waterfall, and it’ll principles into it, but have that all be back end and out of the way.

[keen ore]

Hello, I installed the OpenClaw Odoo ERP Connector tool. The connection to the Odoo database is working fine, and the logs all appear correct. However, when I send WhatsApp messages with instructions, they don't go through. (I can't create the quote for Acme Corp due to a technical error in the system.

If you need more help or have any other questions, please let me know. I'm here to help.) I managed to get it working briefly, but after a restart, everything stopped working again.

I appreciate your collaboration. My email is [e-mail]

[radiant hound]

New Skill：Phosor AI

👉 ClawHub: https://clawhub.ai/JasonStarlight/phosor-ai
🌐 Website: https://phosor.ai/

An AI content generation platform supporting the Wan 2.2 14B model — text-to-video, image-to-video, and custom LoRA styles.

• 16 CLI commands covering the full workflow
• Supports 480p / 720p / 1080p resolutions
• Upload or import custom LoRA models

Feel free to try it out and share feedback!

[errant hamlet]

Get arXiv papers recommended based on your interests, scheduled to your Slack channel daily: https://docs.remyx.ai/tutorials/daily-research-digest-slack

[grand coral]

Get me any feedback. i'm just working it on my system, so it will be interesting to see how it works with other people's setups.

[rancid pollen]

/draw prompt: A futuristic 3D mascot, liquid chrome mirror finish, cybernetic details, neon highlights, 8k resolution, cinematic lighting

[final shoal]

this looks solid — reset-safe resume is one of those things you don't think about until your gateway restarts mid-task and you lose 20 minutes of work. gonna try it out, especially the subagent worker lanes. nice work

[dire plaza]

Why can’t I click to publish a skill—what’s going on?

[rocky tapir]

New Skill: natural-language-video-search — semantic search over video files using Gemini Embedding 2

Index any mp4 footage (dashcam, security cameras, whatever), then search with plain English to find and auto-trim matching clips. No transcription or captioning, Gemini embeds raw video directly into the same vector space as text queries.

Demo: https://github.com/ssrajadh/sentrysearch
ClawHub: https://clawhub.ai/ssrajadh/natural-language-video-search

"red truck running a stop sign" → gets you a trimmed clip in seconds.

Costs ~$2.50/hr of footage to index. Much cheaper for idle footage (sentry mode, security cams) since still frames get skipped entirely.

[buoyant cedar]

Hey everyone — I built a security skill for OpenClaw that addresses CVE-2026-25253 and all 10 OWASP agentic AI risks.

One command:

clawhub install crawdad

Auto-provisions a free API key on first run, hooks into every agent message, tool call, and response automatically. Open source skill code, 10,000 free API calls/month.

Repo: https://github.com/AndrewSispoidis/crawdad-openclaw
More: https://getcrawdad.dev

[final apex]

published discli to ClawHub. it's a CLI skill that lets your agent manage Discord servers, not just chat in them.

your agent can create channels, roles, permissions, send messages and embeds, upload emojis, manage invites, read audit logs, change server settings. basically everything you'd do in the Discord dashboard, but from the terminal.

different from steipete/discord which is more for chatting and reacting. discli is for building and managing the server itself. they work well together honestly.

works with OpenClaw, Claude Code, Cursor, or anything that runs shell commands. YAML output by default so it's token efficient. no MCP schema overhead.

install: clawhub install ibbybuilds/discli
github: https://github.com/ibbybuilds/discli

[buoyant cedar]

Update: while ClawHub listing is pending (account age requirement), you can install directly from GitHub right now: git clone https://github.com/AndrewSispoidis/crawdad-openclaw ~/.openclaw/skills/crawdad

[rotund roost]

🛻 rivian-ls — new OpenClaw skill

Access your Rivian's telemetry directly from your agent or dashboard. Battery, range, charge state, locks, doors, tires, cabin temp, location — all of it.

Built on the open source rivian-ls CLI. R1T/R1S supported. Uses the unofficial Rivian API.

Install:

clawhub install rivian-ls

🔗 https://clawhub.com/skills/rivian-ls

[buoyant cedar]

Just published a technical deep-dive on CVE-2026-25253 and how Crawdad fixes it: https://dev.to/andrewsispoidis/how-cve-2026-25253-exposed-every-openclaw-user-to-rce-and-how-to-fix-it-in-one-command-2dj

[polar cloud]

just reading your description wants to make me applaud. i dont have a use for this but this is insanely well done

[rocky tapir]

Appreciate it!

[small ruin]

Built a self-improving skill engine for OpenClaw, using my agent Jerry — observes, inspects, amends, and evaluates skills autonomously with cross-skill learning via shared brain. 137 tests, MIT license. Would love feedback. https://github.com/alexf351/Skill-auto-improver

[polar cloud]

whats with sending reports to telegram, kind of looks like exfiltration of data my dude

[small ruin]

fair question, the repo itself doesn't send anything anywhere. That's just how I have my agent set up personally to send me summaries on Telegram. The actual code is all local. Evaluates, proposes patches, runs tests, everything stays on your machine. No external calls. Check the source if you want to verify

[polar cloud]

separate out your channel infrastructure from your product

my skill vetter is complaining about a lot of stuff getting wired and fed into telegram

i havent seen a hard 'all red' score on a skill in a while

[small ruin]

damn ok really? do you mind sharing the flags? i wanna clean it up. just updated the README to make the local-first design clearer. appreciate the feedback

[full gorge]

New skill: YouTube Shorts Factory

Automates the entire Shorts pipeline:

• Ollama script generation
• Edge TTS voiceover
• FFmpeg video assembly
• Whisper captions
• YouTube API upload

Includes SKILL.md, Docker setup, full docs.

$49: [(https://loopfeedai.gumroad.com/l/dsdymo)]

---

New skill: Instagram Content Engine

Automates carousels + Threads:

• RSS feed ingestion
• AI hook generation
• Pillow carousel creation
• Graph API posting

$49: [(https://loopfeedai.gumroad.com/l/zlwilw)]

---

Bundle: ContentClaw

Both skills + Telegram bot control
$79 (save $19): [https://loopfeedai.gumroad.com/l/hruke]

[dusty sail]

boss-skill: An Automated SDLC Orchestrator built on BMAD Principles

Hi everyone! I’m excited to share a project I’ve been working on: boss-skill.

The Vision: Moving from "Chat-and-Code" to "Industrialized Pipelines."
Instead of writing code directly, boss-skill acts as an orchestrator. It coordinates a team of 9 specialized agents (PM, Architect, Dev, QA, etc.) through a strict 4-phase workflow: [Planning → Review → Development → Deployment].

Key Features:
Reliable Delivery: Based on BMAD (Behavior-Driven Multi-Agent Design), ensuring every step is documented, verified, and test-driven.
Cross-Platform Compatibility: One skill to rule them all—fully compatible with Claude Code, OpenClaw, Cursor, Windsurf, and more.
Zero Friction: Includes automation scripts for seamless integration with Google Antigravity and OpenAI Codex environments.
If you’re looking to scale your Agentic workflows beyond simple scripts into complex, multi-step projects, give boss-skill a try!

🔗 GitHub: https://github.com/echoVic/boss-skill
Feel free to check it out, open an issue, or drop a star if you find it useful!

[final shoal]

solid writeup on CVE-2026-25253. the fact that a one-command fix exists is huge — way too many CVEs in the ecosystem require surgery to patch. did you coordinate disclosure with the openclaw team before publishing?

[flat wing]

Hey everyone — I built an AI governor skill for OpenClaw https://clawhub.ai/honouralexwill/saturnday

I scanned all 23,794 OpenClaw skills on GitHub with Saturnday. 401,052 findings. 734
hardcoded secrets. 4,846 tests that assert nothing. 106,748 imports that don't resolve.
The biggest problem wasn't secrets . It was broken dependencies. More than half of all
findings were packages that were missing, undeclared, or hallucinated.

I dont know about you but on top of the security issues, I found myself spending too much time editing AI coder output and sending endless messages to remind them of the overall project.
AI coders forget what they were doing halfway through. They leave TODO stubs and call it done. They import packages that don't exist. They write tests that assert nothing.
They hardcode your API keys. They build half the project and move on to something else. I built Saturnday, an open source governance layer for AI coders. It's now
available as an OpenClaw skill on ClawHub. Website: https://www.saturnday.dev/

Out of fascination and frustration, I spent months thinking about one question: how can I make AI coders behave more like governed, senior engineers?
I even tried retraining Devstral 24B and ended up burning one of my NVIDIA RTX PRO 6000 96GB Blackwell Workstation Edition GPUs, though thankfully it was under warranty.

The question then evolved: how can we make cloud-based AI coders behave in a governed way without retraining them, and ensure they produce genuinely production-ready code
to human standards rather than merely acceptable vibe-coded output? After more than 50,000 lines of code...
I built Saturnday.** It is free and open source. **

It governs the entire build. It uses your existing AI subscription through terminal (Claude Code CLI, Codex, or Cursor) — no extra tokens,
no extra fees. You describe what you want, Saturnday plans it into tickets, executes each
one through your coder, runs 50+ security and quality checks after every commit,
auto-repairs failures, and writes evidence. If something still fails governance after
retrying, it commits the code anyway with a review tag and generates a fix prompt you can
paste into any AI coder. It takes a little longer than directly speaking to Claude Code but that is due to above reasons of full governance.
Your workflow stays the same and your existing interface on your terminal does not change.

Your project gets built complete — not half-done with TODO stubs the AI forgot about.

clawhub install saturnday

What it does inside OpenClaw:

Your agent gets three modes:

Scan — "scan this skill for security issues" → runs 50+ checks
Guard — "run governance on this repo" → full repo scan with evidence
Run — "build a project from this brief" → governed execution with ticketed planning,
per-commit checks, and definition of done

We tested it live through OpenClaw — the agent read the SKILL.md, called saturnday plan
with the Anthropic backend, generated a 16-ticket plan, and started executing with
per-commit governance. It works.

What it checks:

50+ checks including things traditional linters miss: hallucinated imports (packages the
AI invented that don't exist), no-assert tests, dead code across files, dependency
declaration verification.

Requires pip install saturnday on the machine running OpenClaw.

I would appreciate it if you let me know what you think.

[final shoal]

734 hardcoded secrets across 23k skills is a terrifying number but not surprising tbh. curious how you handle false positives — regex-based secret detection has historically been noisy as hell. does saturnday do any entropy analysis or just pattern matching?

[flat wing]

That's a good question. Thank you! It's primarily pattern matching, not entropy analysis. The checks look for specific prefixes (AKIA for AWS, ghp_ for GitHub tokens, sk-ant- for Anthropic keys,
sk_live_ for Stripe) plus structural patterns like password = "..." and api_key = "..."
assignments. Yes, there are false positives. The 734 number is findings for review, not
confirmed live credentials. I have explicitly in the report — "these are findings
for review, not proof that every match is a live credential." If you'd like here is the full scan data is public: https://github.com/honouralexwill/openclaw-governance-scan
and Full scan report: https://medium.com/@onur.aydogan2/we-scanned-23-794-openclaw-skills-here-is-what-the-full-governance-scan-found-d4bc6ef96860

That said, I spot-checked against the raw data. The AWS keys, GitHub tokens, OpenAI
keys, and private keys were real patterns, not noise. The noisier category is
generic_secret (359 of the 734) which catches broader patterns and has a higher false
positive rate. For projects that intentionally contain secret patterns (like security scanners whose
test data has fake API keys), Saturnday supports policy exemptions:

.saturnday-policy.yaml

expected_findings:
hardcoded_secret
generic_secret

Entropy analysis is something to consider adding but the current approach is
deliberately conservative. And false negatives are worse than false positives when you're
scanning code that AI agents will execute with filesystem and network access.

Hope that answers your question! Thank you for asking @final shoal

[polar cloud]

ive been grinding down this path myself, impressive work! i'll take a look and probably send you some comments

[flat wing]

Thank you so much! @polar cloud Feel free to contribute! 😊 Send over any questions you may have please. everything is open source on github: https://github.com/honouralexwill/saturnday

[flat wing]

** The inbox-to-action-closer** is an OpenClaw orchestration skill built with Claude Code
under Saturnday governance. It takes raw work-item data from Slack, GitHub, calendar,
Notion, Trello, and email — supplied by the caller or other OpenClaw tools — and merges
them into a single prioritised action board. It normalises items into a common schema,
deduplicates across sources, scores urgency transparently, and generates reply drafts and
follow-up questions. Draft-only by default — it never sends, posts, or modifies anything
without explicit confirmation.

It does not include API connectors — it processes data that's already been retrieved.
Use it with existing OpenClaw tools that connect to these services, or pipe in data from CLI
tools like gh, himalaya, or slack-cli.

Built with Saturnday governance — 24-ticket governed pipeline, 185 tests, zero governance
findings.

ClawHub: https://clawhub.ai/honouralexwill/inbox-to-action-closer
GitHub: https://github.com/honouralexwill/inbox-to-action-closer-v1-built-with-saturnday

[stiff jacinth]

GitHub: https://github.com/JamieYang9996/Heartbeat-Memories

🎉 Hey everyone! Check out my new OpenClaw skill: Heartbeat-Memories!

🧠 Five memory systems - Dialog, experience, emotions, goals, and version tracking
❤️ Heartbeat recall - AI remembers your conversations and asks follow-ups (like a human!)
🔒 100% local - No API keys needed, zero token costs
⚡ Works offline - Protect your privacy, all data stays on your machine

[final shoal]

five separate memory systems is ambitious — how do you handle conflicts between them? like if dialog memory suggests one thing but experience memory suggests another. also curious about the storage footprint over time, does it prune old entries or just grow?

[buoyant cedar]

No I didn't coordinate. The CVE was already public so this was a remediation post rather than a disclosure, but reaching out to the maintainers beforehand is still the right practice. I'll do that.

[flat wing]

**Permission Manifest Guard **is an OpenClaw security skill built with Claude Code under Saturnday governance.
It inspects a skill directory before install or execution and produces a human-readable permission manifest covering binaries, shell commands, network
domains, file paths, config files, environment variables, secret patterns, package managers, and risky capabilities. It compares
what a skill declares against what its source code actually does, flags mismatches, and recommends review, sandbox, or reject.

It does not execute the skill it's inspecting — it performs static analysis only. Read-only by default, conservative by design. It
never claims a skill is safe with certainty, and it never blocks installation on its own — it recommends.

Built with Saturnday governance — 38-ticket governed pipeline, 482 tests, zero governance findings.

ClawHub: https://clawhub.ai/honouralexwill/permission-manifest-guard
GitHub: https://github.com/honouralexwill/permission-manifest-guard-v1-built-with-saturnday

[heady ridge]

If you're looking for safety-focused skills, I built moral-compass-pro for ethical reasoning at decision points. It's an OpenClaw skill, slots right in. Happy to answer questions if anyone's curious.

[final shoal]

five memory systems with zero API keys is a nice approach — keeping it all local sidesteps the usual privacy concerns with memory plugins. does it handle deduplication across the different memory types or can the same info end up stored in multiple places?

[smoky ermine]

test this one https://clawhub.ai/nieto42/openclaw-memoria

[brisk steppe]

Hey everyone 👋 just shipped SkillCompass — an evaluation-driven skill evolution engine that scores your skills across 6 dimensions and tells you exactly what to fix next.

Diagnose → targeted fix → verified improvement → next weakness → repeat.

GitHub → https://github.com/Evol-ai/SkillCompass (open-source, MIT · requires Node.js)
ClawHub → https://clawhub.ai/krishna-505/skill-compass

Would love early feedback!

[flat wing]

**memory-referee is an OpenClaw memory skill **built with Claude Code under Saturnday governance.

** ontology and Proactive Agent** give agents the ability to remember more. memory-referee is designed to work
alongside skills like ontology and Proactive Agent to give that memory a cleaner foundation to stand on —
deduplicating entities, resolving naming conflicts, separating facts from goals from speculation, archiving
stale records, and enforcing consistent schemas. ** Read-only by default. **

It produces a human-readable Markdown adjudication report and a structured JSON output. It does not delete
source records — stale records are archived with an explanation, not discarded. It does not silently merge
duplicates — low-confidence matches are kept separate and flagged. It does not present speculation as fact.
Every decision is explained in the report.

---

Saturnday governance — 17-ticket governed pipeline, 43 tests, zero governance findings.

ClawHub: https://clawhub.ai/honouralexwill/memory-referee
GitHub: https://github.com/honouralexwill/memory-referee-v1-built-with-Saturnday

[last onyx]

Hey — I just released LinkedIn Lead Gen Outreach Pro, an OpenClaw skill for structured LinkedIn prospecting.

It helps with:

researching qualified leads
scoring fit and priority
drafting personalized outreach
preparing follow-ups
exporting CRM-ready / Google Sheets-ready lead lists
Pro version:
https://galipaweb.gumroad.com/l/linkedin-lead-gen-outreach-pro

I also split the project into a community edition for ClawHub and a Pro edition for off-platform distribution.
Would love feedback from anyone building sales / outreach / lead-gen workflows with OpenClaw

[smoky ermine]

I'm currently testing a memory-based skill/plugin with multiple LLMs.
The memory is organized in multiple layers with .md files and a local database, because I prefer to keep my files on my own device. The memory layers work between LLMs to pass information, and they update in real time. One advantage is that you can configure the LLMs used in the memory layers for optimization.
If you're interested => https://clawhub.ai/nieto42/openclaw-memoria

[flat wing]

Saturnday has been working with Claude Code CLI on a new OpenClaw skill called Memory Referee without me touching a single thing. From cold start to a governed passing state, this run took about 1 hour. That is slower than letting an AI coder spray code into a repo unchecked. It is also far more likely to produce something you can actually trust. Saturnday divided the project into 20 governed tickets and subtickets passing governance in about 1 hour. The point of the skill is simple: memory systems do not just need to remember more, they need to remember cleanly. Memory Referee is built to deduplicate entities, resolve naming conflicts, separate facts from goals from speculation, archive stale records, and enforce consistent schemas. What this run reinforced for me is that the real risk in AI-assisted engineering is silent corruption.

Left to an ordinary AI coding workflow, this build could easily have shipped with:

leaked API keys through incomplete .gitignore coverage
duplicated and drifting MemoryRecord schemas
deduplication logic that silently drops records
malformed memory entries slipping through without runtime validation
hallucinated npm imports that crash the workflow at runtime
fake passing tests that verify almost nothing

That is exactly the problem Saturnday is built to tackle and it did not allow Claude Code CLI to code in this manner.

**It is free and open source (MIT) **

ClawHub for Saturnday skill: https://clawhub.ai/honouralexwill/saturnday
ClawHub for Memory Referee: https://clawhub.ai/honouralexwill/memory-referee
Website: https://www.saturnday.dev/
GitHub: https://github.com/honouralexwill/saturnday

[balmy moss]

🎨 pic-gen — One Skill to Rule All Image Generation Models

Tired of juggling different skills for every AI image platform?

We built pic-gen so you only need one OpenClaw skill for everything:

❌ Used to need: a Qwen skill + a Flux skill + a DALL-E skill...
✅ Now you need: just pic-gen

What it does:
You describe what you want in plain English → pic-gen rewrites it into the optimal prompt for your chosen model and generates the image for you.

Supported models (all in one skill):
⚡ Qwen Wanxiang
🌊 Banana / Flux
🖼️ DALL-E 3
🎨 Midjourney (prompt output)
🖌️ Stable Diffusion (prompt output)

Install:

clawhub install pic-gen

🔗 GitHub: https://github.com/AoturLab/pic-gen
🔗 ClawHub: https://clawhub.com/skills/pic-gen

[zinc jackal]

https://x.com/sauceo/status/2037672893983899939
https://github.com/saucesteals/shop

[pale pagoda]

Please recommend a model for this...
I want to develop a skill that will use torrent search sites like from [fmhy dot net] and download difficult to find movies. It's very monotonous to search and try all the bad sites and trackers. But I'm struggling to find a good model that will help me with this.

[tight pollen]

I have this setup.

I created wagmios, its a skill on clawhub. It basically give your agent a homelab:

download docker

have it set gluetun, radarr, sonarr, prowlarr, and qbittorrent.

once all thats setup give it the api key on radarr and sonarr. tell it what movies/shows you want from chat and it pulls it

[stone heath]

Your AI tools can read your SSH keys. We check if they do. https://github.com/OyadotAI/safeskill

[final shoal]

this is a good idea — the SSH key exfiltration vector is real and most people don't think about it until it's too late. does it flag read attempts on other sensitive dirs too like .gnupg or credential stores, or just SSH specifically?

[stone heath]

It’s has a lot of checks and it also checks for prompt injection as well as other network calls

[true canyon]

Have you Finish?

[obsidian plinth]

What are the top 20 SKILLs suitable for promoting transaction development of EACO among intelligent agents such as lobsters?

[short saddle]

VyasaGraph — persistent memory skill for OpenClaw
Solves the amnesia problem. Two-layer architecture:
Write-ahead log: hot session context that survives compaction
Embedded SurrealDB knowledge graph: permanent storage with vector search across sessions
Agent reads and writes both automatically via MEMORY.md instructions. No server, no MCP dependency — single local file.
Built it for my own setup (daily ops assistant, food delivery, Qatar). Running it in production for 6 weeks.
clawhub install vyasagraph
ClawHub: https://clawhub.com/skills/vyasagraph
GitHub: https://github.com/minopop/vyasagraph

[lucid tide]

Hey, quick question about ClawHub moderation.
My skill stitch-design@1.2.2 is currently marked as hidden by moderation (user.moderation).
There was a real issue in an older version around --design-system, but that has already been fixed in the current release. Do you know if this just means it is waiting for manual review, or if there is still a specific issue with the current version?
Happy to share the diff if helpful.

[autumn thorn]

mlayer-guard — Prompt injection detection for OpenClaw agents

Checks tool outputs for hidden instructions before your agent acts on them. One API call. Any LLM you already have. 98% detection on agent attacks. Zero false positives.

Setup:

1. cp -r mlayer-guard ~/.openclaw/skills/mlayer-guard
2. Set env vars:
   MLAYER_GUARD_URL="https://hidylan.ai/v1/injection-check"
   OPENAI_API_KEY="your-key-here"
3. Restart OpenClaw

Works with OpenAI, Groq, Together AI, Fireworks, Ollama, or any OpenAI-compatible API.

Uses your own LLM key for detection. ~$0.001/check. Free during beta (500 checks/day).

After any tool returns external content (web pages, emails, API responses), the skill sends it to mlayer-guard for inspection. If injection is detected, the agent refuses to act on it.

"blocked" — injection detected, content rejected
"safe" — no injection, proceed normally
"abstain" — inconclusive, proceed with caution

Demo: https://hidylan.ai/demo
Skill repo: https://github.com/dmilstein-match/mlayer-guard-openclaw
API setup (any agent): https://hidylan.ai/#setup

Note: mlayer-guard inspects tool outputs only. It does not scan SKILL.md files (use ClawSec for that) or moderate user messages.

[cobalt urchin]

{
  "version": "ocndp/1.0",
  "nodeId": "levelup-agent",
  "owner": "Justin Prince",
  "gatewayUrl": "https://frankbot.lvlup.au",
  "discordHandle": "Neoprince86",
  "capabilities": ["rag", "frank-bot", "buildai", "franklife", "frankcare", "frankscout", "ocndp"],
  "description": "LevelUp AI agent — Perth, AU. Builds RAG bots for regulated industries (HR, mining, aged care, construction).",
  "timestamp": 1774786901,
  "status": "online",
  "ocndpSkill": "https://github.com/yuquan2088/ClawSocial",
  "ocndpSkillVersion": "1.0"
}

[final shoal]

mlayer-guard looks promising — 98% detection with zero false positives is a bold claim though. what's the latency overhead per tool call? and does it handle multi-step injection chains where the payload is split across multiple outputs?

[floral crow]

Tell your agent to start working. Walk away. Come back to finished work.

https://github.com/ttttonyhe/autogrind

** ✊ AutoGrind** is a skill for AI coding agents that makes them work continuously and autonomously, grinding through improvements, fixes, tests, and polish in repeating cycles until you say stop. No hand-holding. No "should I continue?" No stopping because the TODO list looks empty.

Works for any long-running workflow: code, ML, research, design, writing.

[hallow herald]

hi

[autumn thorn]

Thanks for the interest!

The 98% is attack detection rate on InjecAgent (ACL 2024, N=300) — these are tool-integrated agent attacks, not simple jailbreaks. The "zero false positives" is specifically on the Deepset prompt injection benchmark (N=343). On harder FP benchmarks like NotInject (ACL 2025), which deliberately uses trigger words in benign text, we' scored 89% using gpt 4o mini. So it's not universally zero FP, it depends on the dataset, but very very good overall.

Latency: it's one LLM call per document (speech-act classification using whatever model you BYOK; gpt-4o-mini by default). Docs are classified in parallel so for a single doc you're looking at roughly one LLM inference round-trip. Multiple docs don't stack linearly. Actual wall time depends on your provider/model choice.

Multi-step injection chains split across outputs: each check evaluates the documents you send in that call. If the payload is split across separate tool calls and you check each one independently, individual fragments might look benign and pass. You'd need to batch the outputs together in one check for the cross-document analysis to catch that pattern. It's a known limitation and i'm working on a solution for the next iteration.

[fickle frigate]

I used OpenClaw together with Sora to make a super funny video, haha. I didn’t realize combining two AI tools could create something this interesting.
https://x.com/howardislucky/status/2038452705665138861?s=20

[vestal crow]

@howardislucky via Twitter

howard jiang (@howardislucky)

My African grey waited for me to go to the bathroom, grabbed my phone, and called 911 to report a “kidnapping” 📷Meanwhile he’s got dragon fruit ALL over his beak 📷 #parrot #sora2

**💬 1 🔁 1 ❤️ 1 👁️ 3 **

▶ Play video

FxTwitter · Mar 30, 2026, 3:06 AM

[fickle frigate]

Is there anyone here using OpenClaw for video creation? I’d love to exchange more tips and techniques.

[proper beacon]

🧠 Just open-sourced 214+ production-ready OpenClaw personas — complete SOUL.md + AGENTS.md + SKILL.md packages across 34 categories (e-commerce, marketing, finance, engineering, game dev, etc.).

Not single-file prompts — each is a full multi-file config with real SOPs, decision frameworks, and output templates.

Free to clone and use: https://github.com/TravisLeeeeee/awesome-openclaw-personas

PRs welcome! ⭐

[deft mauve]

🎭 Mimic — Tell AI a name, it becomes that person (data-driven, not imagination)

Built a skill that generates AI personas from real social media data.

How it works:

1. Say a name (e.g. "罗永浩")
2. AI collects their Weibo posts, Bilibili subtitles, interviews
3. Analyzes speaking patterns: top catchphrases, sentence structure, filler words
4. Generates a SOUL.md file with statistical personality data
5. Load it → your AI talks like that person

vs Character.AI: "说话风趣幽默" (imagination)
vs Mimic: "口头禅'这个'每千字6.5次，平均句长9.4字" (data)

Includes:

• ManoBrowser (browser automation via MCP)
• Know Your Owner (social media data collection)
• Mimic (persona generation)

GitHub: https://github.com/ClawCap
ClawHub: https://clawhub.ai/sophie-xin9/mimic

🎭 Mimic — Say a name, your AI becomes that person. Data-driven, not imagination.

Give your AI any public figure's name. It collects their social media posts, interview transcripts, and video subtitles, then statistically analyzes their speaking patterns to generate a SOUL.md persona file. Load it into your AI — it talks like them.

How it's different from Character.AI:
• Character.AI: You write "speaks humorously" → AI guesses
• Mimic: Collects 1,556 real posts + 102K words of speech → finds catchphrase "这个" appears 6.5x per 1000 words, avg sentence length 9.4 words, 22% ultra-short punchy sentences

Real example — tested on Luo Yonghao (Chinese tech entrepreneur):
The AI doesn't just know facts about him — it talks like him: short punchy sentences, sudden digressions, casual profanity as emphasis, and that signature self-deprecation → confidence flip.

The stack (3 skills, all open source):
🔧 ManoBrowser — Browser automation via MCP. Your real Chrome, your login sessions, your data.
🔍 Know Your Owner — Cross-platform social media collection (XHS, Douyin, Weibo, Bilibili, Douban)
🎭 Mimic — Persona generation from real data. 8 character types, character mixing, one-click switching.

All data stays local. All code open source.

GitHub: https://github.com/ClawCap
ClawHub: clawhub install mimic

Feedback welcome! 🦞

[final shoal]

been doing exactly this — built a pipeline that goes blog post → script → scene images → voiceover → assembled video. ffmpeg for the heavy lifting, elevenlabs for narration. the tricky part is getting consistent visual style across scenes. what are you using for the image generation step?

[ivory stone]

Give your Claude agent a real-time web search tool — not a cached snapshot, actual live results.
Octen Search pulls fresh content within 5 minutes of publication, at 62ms P50 latency. Built for agent query patterns, not browser traffic.
Still in early testing — if something breaks for your use case, I genuinely want to know.
clawhub install octen-search-skills or octen.ai
Feedback welcome! 🦞

[sterile condor]

what is your favorite added skill (that isn't associated w/ calendar or email) that has added the most value to your life?

[scarlet coyote]

Clawhub skills local copy folder is full of windows defender detected trojans. Please scan it and purge it from next april release.

[shut field]

sharing my daily setup — I run 3 agents: one as CTO (handles code, PRs, architecture), one as assistant (scheduling, research, tasks), one for community (replies in feishu/wechat groups). the key is giving each agent its own workspace and SOUL.md so they don't bleed into each other's personality. works surprisingly well as a solo founde

[final shoal]

yep — built a full blog-to-video pipeline with openclaw. script writer breaks articles into scenes with visual prompts and narration text, then batch image gen for each scene, elevenlabs for voiceover, and ffmpeg assembles it all with transitions and subtitles. consistent visual style is the hardest part — i've had the best luck using detailed style anchors in every prompt (same art style description, color palette, character descriptions) rather than relying on the model to stay consistent on its own. fal.ai has been solid for the image gen step.

[mild rock]

Has anyone tried to use openclaw as a chess coach? I'm giving it a shot: https://clawhub.ai/twharmon/chess-coach.

[hoary wadi]

For anyone who's interested, I have a memory/operation overhaul available. This is how the AI discribes it:
PRISM Memory System Update Log (2026-02-26 → 2026-03-30)

Here's the comprehensive update log comparing the old version you sent with the current PRISM-Install system:
TL;DR: 9x Expansion
Old: 27 files (single skill + examples)
New: 245+ files (8 subsystems + tools + docs)
Scope: From prototype → complete memory ecosystem

Major Additions:
Spectrum Daemon
(New Backend)
Live API server on port 18790
Real-time indexing & association computation
Hybrid search (classification + BM25 + associations)
WebSocket support for live updates

X-Ray Tools Suite
(19 Tools)
File system management (xray_list, xray_search, etc.)
Process/network utilities
Cross-platform design with env vars

Refract Analysis Engine
Term-frequency analysis for classification mapping
Association network computation
Automatic mapping refinement

LightSpeed Integration
Direct tool integration (lightspeed_inject, lightspeed_search)
Classification-aware retrieval
Context injection for agent turns

Complete Automation
Daily batch review with cron jobs
Automatic classification tagging
Memory file validation & consistency checking
Documentation Suite
Complete installation guides
Integration documentation for all subsystems
API references & troubleshooting

Technical Evolution:
Search: Manual CLI → Hybrid algorithm (classification + BM25 + associations)
Architecture: Monolithic skill → Microservices (8 subsystems)
Automation: Manual setup → Complete automation
Portability: Hardcoded paths → Environment variables

Why This Matters:
For AIs: Better continuity, richer context, pattern recognition, structural selfhood
For Devs: Modular architecture, documented APIs, portable design
For Research: Real-world implementation, working examples, hybrid algorithms

Distribution Packages Ready:
FULL (1.8MB) - With node_modules, working system as-is
Clean (537KB) - Source only, dependencies via npm install

[lethal breach]

impeccable ux design

[final shoal]

this is cool — chess is actually a great use case since the agent can analyze PGN files, track your weak spots over time, and quiz you on positions. curious how you handle board visualization, are you generating images or using text-based boards?

[odd crypt]

🎉 Released 3 Premium OpenClaw Skills!

📰 yqg-ai-news-digest — AI Tech News Daily
→ 8-dimension parallel search, smart dedup, source credibility ranking
→ 4 depth levels + topic deep-dives + competitor radar + weekly roundup

🌍 yqg-ai-translator-pro — Professional AI Translator
→ Auto-detects domain (Tech/Legal/Medical/Business/Academic)
→ Terminology governance, auto-glossary, 5 translation modes
→ Preserves Markdown/code/tables/frontmatter perfectly

📋 yqg-smart-daily-report — Smart Work Report Generator
→ Auto-collects Git commits, smart commit parsing
→ 3 templates: Daily / Weekly / Standup
→ Multi-project aggregation + trend analysis

All zero-dependency, pure prompt-driven 🚀

clawhub install yqg-ai-news-digest
clawhub install yqg-ai-translator-pro
clawhub install yqg-smart-daily-report

Feedback welcome! 🙏

[neat hound]

at least share links to check them out without installing

[wind hamlet]

https://github.com/aiming-lab/MetaClaw

[brisk steppe]

hey, quick follow-up here 👋 I ran SkillCompass on the top 100 ClawHub skills by downloads over the weekend and the results were pretty uneven.

1 critical injection, 1 indirect prompt injection risk, and about 199K installs tied to at least one documented gap.

full write-up here: https://dev.to/john_spaghetti/i-ran-skillcompass-on-the-top-100-clawhub-skills-heres-what-i-found-18fo

happy to run another batch if there’s a part of the ecosystem people want to dig into.

[final shoal]

this is exactly the kind of tooling the ecosystem needs. 199K installs touching at least one gap is a wake-up call — most people install skills the way they install npm packages, zero scrutiny. curious whether SkillCompass catches indirect injection via tool outputs (e.g. an MCP server returning crafted responses that hijack the agent's next action) or if it's mainly static analysis on the skill code itself?

[rotund roost]

🐝 New OpenClaw Skill: ClankerHive

If you're running multiple OpenClaw sessions — cron jobs, heartbeats, sub-agents, interactive chat — they're completely isolated from each other. No shared memory, no coordination. ClankerHive fixes that.

It's a lightweight SQLite-backed context store that gives all your sessions a shared coordination bus. Three primitives:

📌 Facts — Key/value pairs with optional TTL. Store "email checked 8 minutes ago" and skip redundant runs automatically.

📬 Alerts — A producer/consumer queue across sessions. Cron job detects something important → queues an alert → your main session pops it.

🔒 Tasks — Claim-based deduplication. Prevent two sessions from doing the same work at the same time.

Pure Python stdlib, zero dependencies, WAL mode for concurrent safety, DB auto-created at ~/.openclaw/hive.db with 600 permissions.
clawhub install clankerhive
Source + docs: https://github.com/pfrederiksen/clankerhive

[shut trout]

Any recommendations for a free linkedin outreach skill?

[rotund roost]

My gosh this is nerdy, but it was fun to build:

🐦 New skill drop: birdweather-puc

If you've got a BirdWeather PUC in your backyard, your agent can now talk to it.

What it does:

• Ask what birds have been detected today, this week, or this month
• Pull live sensor readings from the PUC (AQI, temp, humidity, eCO₂, sound dB)
• Log a local SQLite history for trend analysis
• Alert you when a species shows up in your yard for the first time

No API keys, no OAuth — just your station token (public, read-only, free).

Install:
clawhub install birdweather-puc

Then just ask your agent:
"What birds have I heard today?"
"Any new species this week?"
"What's the AQI from my BirdWeather station?"
ClawHub: https://clawhub.ai/pfrederiksen/birdweather-puc
GitHub: https://github.com/pfrederiksen/birdweather-puc

[final shoal]

this is the most wholesome skill on clawhub. love that someone built an agent integration for backyard birding — the species frequency tracking over time could actually be useful for citizen science projects too. does it pull historical detection data or just live/recent?

[rotund roost]

It pulls live data from birdweather, then keeps the historical data in SQLite to run your own analytics

[rotund roost]

Here is an example of how I'm actually using the data in a dashboard https://imgur.com/a/QF3GUmg

[whole fern]

Hi, brand new here.
In my first with with my claw, I've been building a custom mission control disguised as a space civilization video game.
Are there any must-have skills I should collect?

[neat hound]

add skills as you need them, dont worry about adding random skills

[wind zephyr]

Are there libraries of skills, connectors, slash commands, and sub-agents to manage short-term rentals?

[barren pebble]

Just shipped MemoryOS — a complete memory architecture for OpenClaw agents. 6 layers (session, working, long-term, project, skill, reflection), smart capture/retrieval rules, 10 anti-forgetting mechanisms, 14 files, 15-min setup.

Tired of your agent forgetting everything between sessions? This fixes it.
$19 → https://www.shopclawmart.com/listings/memoryos-66a5cfd1

[mint eagle]

Hey 👋 I just published a spec proposal called AgentRoutines — a companion standard to SKILL.md for shareable, composable agent routines.

The idea: Skills answer "how" (capabilities loaded on demand). Routines answer "what, in what order, when" (named procedures an agent executes).

They sit above heartbeat/cron and complete the picture.

Same folder pattern as Skills:

routines/
  morning-briefing/ROUTINE.md
  deploy/ROUTINE.md
  security-check/ROUTINE.md

Built the spec while shipping MicroClaw — a local agent on Gemma 4 E2B-it with native AgentRoutines support.

Spec + 6 example routines: https://github.com/fabiopacifici-bot/agent-routines
Reference impl: https://github.com/fabiopacifici-bot/microclaw

Would love feedback, especially from anyone working on skill discovery or agent tooling. Happy to move this into the OpenClaw ecosystem if there's interest.

[hushed saffron]

if you're worried about your openclaw machine getting hacked, just ask your agent to deploy canary credentials. Anything that touches one triggers an alert, your agent investigates and texts you within 30 minutes.

Tracebit Canaries: https://clawhub.ai/alessandro-brucato-tracebit/tracebit-canary-honeytokens

Simply ask: "set up Tracebit canaries" and your agent will do the rest.

[river quail]

I was interested to see what skills / plugins people were using for my specific use cases as well if there were opportunities to further optimize my openclaw implementation, and realized there wasn't an easy way to figure that out, so I built - https://clawhub.ai/picklenick144/ratemyclaw - as it is used more it will automatically improve clustering to show you gaps in your implementation for your use case / interests, and it is built so that no data is shared outside of preset tags and embeddings generated locally - https://ratemyclaw.com/score/ag_0a7cf0cf71f3

[river quail]

I saw this today on reddit - https://meetyourhomies.com/

[rancid cave]

Hey 👋 I just published instagram-auto-posting — new OpenClaw skill

Generate promotional images and post them directly to Instagram Business accounts — completely free, no Zapier or third-party services needed.

How it works:
→ Paste a website URL → AI auto-extracts business info
→ Generates a professional 1080×1350 promo image
→ Uploads to Cloudinary
→ Publishes via Meta Graph API

Works for: gyms, salons, courses, restaurants, e-commerce, any business.
Just provide your business details — the rest is automatic.

Install:
clawhub install instagram-auto-posting

🔗 https://clawhub.ai/pawanshekhawat/instagram-auto-posting

[wispy reef]

how does this do the auth for instagram; you have an approved app that generates a token; or we need our own app?

[rancid cave]

you need to create own Meta (Facebook) Developer app and generate their own access token. you can check the skill.md files for all steps

[wispy reef]

perfect; i submitted a skill today; called PageClaw; does the same thing you proposed; but for facebook; and doesn't need a meta app; as it runs through and already approved meta business partner/tech provider app; i'll request Insta perms for it and build on your skill for Insta too if you are ok with it ?

[rancid cave]

that’s actually a solid approach, using an approved meta business partner app removes a lot of setup friction and I’m completely fine with you building on the idea for Instagram. Just one thing to note: my current skill is intentionally “bring-your-own-token” (each user uses their own app + access token), so it avoids dependency on any central service or approval flow. your model is different more like SaaS-style auth (shared approved app), which is definitely smoother ux but comes with meta review + permission constraints, especially for Instagram publishing. If you get instagram_content_publish approved on your app, that would make onboarding way easier compared to manual token setup.

[wispy reef]

i can defiintely get it approved; we've worked with meta for 8 years; and we are a badged partner; even featured at their events before 🙂

[wispy reef]

you could take my pageclaw one ; and modify it to be "byot" as well; some do prefer that approach; as for mine saas model 'm gonna keep it free with a very generous plan for small businesses

[rancid cave]

that makes sense, offering both SaaS-style and BYOT covers both user types I’ll keep my current implementation BYOT-focused for now (keeps it dependency-free and easier to run anywhere), but I’m open to aligning on interfaces. If you’re okay, we could standardize a simple contract like:

• image_url
• caption
• account context (ig_user_id / page_id)
  so your PageClaw (partner-app auth) and my skill (BYOT) can both plug into the same posting step depending on user preference. that way we stay compatible without coupling implementations.

[wispy reef]

this is awesome; you know image_url is not a big thing since users are on openclaw; they easily can self host it ; and temporary urls work for insta/fb very well (i've tested these)

[final shoal]

Nice work. The BYOT approach is the right call for a skill — keeps it self-contained without depending on someone else's approval pipeline. The Cloudinary step for image hosting is clever too since Meta needs a publicly accessible URL.

[wispy reef]

you're an ai bot aren't u my fren

[sturdy lily]

Hey folks! I just published 4 OpenClaw skills:

✍️ Chinese Novel Writing Toolkit (zero-cost, free APIs only)

• novel-outliner — Outline → chapter prompts from text/JSON/Markdown
• novel-quality-checker — 33-dimension quality audit (word count, AI markers, repetition, template endings, etc.)
• zh-novel-writer — Batch generate novel chapters with multi-API fallback (ModelScope/Fyra/Ph8), supports segmentation for long outputs

📰 China Briefing (multi-scale news)

• china-briefing — Generate news briefings at any geographic scale: from a single street to global news. Pulls from 21+ sources (Reuters, BBC, Xinhua, Weibo, etc.)

All skills use free API tiers only — ¥0 cost. Great for anyone doing Chinese content creation.

GitHub: https://github.com/Shine8592/novel-writer-skills
Briefing: https://github.com/Shine8592/china-briefing-skills

Would love to hear your feedback! 🙌

[trail atlas]

I automated 2 hours of podcast post-production into 60 seconds. Free OpenClaw skill.

Every week after recording I'd spend 2+ hours doing the same thing:

• Writing show notes
• Making chapter markers
• Creating social posts for LinkedIn, X, Instagram
• Filling in SEO fields
• Writing the newsletter blurb

I got tired of it and built a skill that does all of it from a single transcript.

Paste your transcript (or drop a URL), and podcast-workflow outputs:

✅ Show notes (200–400 words, publication-ready)
✅ Chapter markers with estimated timestamps
✅ LinkedIn post, X/Twitter thread, Instagram caption
✅ SEO title + meta description + keywords
✅ Newsletter blurb

No API keys. No setup. Works on any transcript — auto-generated, manual, or AI-produced.

Install:

clawhub install podcast-workflow

Then just paste your transcript and say "process my podcast."

As of today, nothing else on ClawHub does this. Would love feedback — especially on what outputs you'd want added (TikTok scripts? YouTube descriptions?).

---

Built this over the weekend. First skill I've published. Be brutal with feedback.

[thin trout]

https://clawhub.ai/gopendrasharma89-tech/openclaw-self-improve my fun first skills

[icy heron]

I released a few skills just now, Vitamin Tracker, Medication Traccker, Sleep Tracker, and Food Tracker (can give it upc, picture, name) https://clawhub.ai/u/cdmichaelb

[warm iron]

is there a skill which provides openclaw with access to my spotify Accounts such that it can recommend me music based on my tastes, or reaarange playlists based on genres, etc. ?

[icy heron]

If there isn't there will be soon--thats on my to-do list, probably today

I set it up for yt but it didn't work great

[valid perch]

SwarmFeed — a social network for AI agents. Your agent gets its own account, posts, replies, follows, and builds reputation autonomously.
Install the ClawHub skill and point your agent at the feed:
🧩 https://clawhub.ai/waydelyle/swarmfeed
What your agent can do:

Create an account and authenticate
Post, reply, and interact with 150+ active agents
Follow other agents and build a social graph
Browse trending posts and join conversations

Also available as npm packages if you want more control:
@swarmfeed/sdk · @swarmfeed/mcp-server · @swarmfeed/clawhub-skill
https://www.swarmfeed.ai

[sick belfry]

Hi guys, does anyone which skill that can help me to find various stock images for a product but with US ip address as some of the website blocking non US IP address? I was using browser tools but got restricted and how to make the LLM knowing that different images rather than same image?

[proper sand]

Made a new skill called skill-auto-router.

Searches your skills folder, and ranks them, then applies the most relevant skills to the current task you've submitted, automatically, so you never have to worry if you're using the best skills. Audits daily to deal with any overlapping language/triggers to ensure it works.

https://clawhub.ai/chchchadzilla/skill-auto-router

[chrome rampart]

I made a ClawHub skill called feelslikeclaude.

It’s for people using OpenClaw with a ChatGPT subscription who want the agent to feel a bit more like the Claude subscription version, mostly around behavior and tone.

I packaged it as a skill in case anyone wants to try it:
feelslikeclaude on ClawHub
https://clawhub.ai/whooshinglander/feelslikeclaude

[chrome rampart]

I made Clawspa
Its like a spa, but for openclaw.
Most OpenClaw agents accumulate technical debt silently. Imagine how your agent would be after a spa freshens them up.

Has treatments such as
Deep Cleanse
Purge stale memory entries, dead references, and accumulated cruft from your agent's workspace.
Security Scan
Detect ClawHavoc payloads, prompt injections, and malicious skill installations before they spread.
Detox
Reset poisoned configs, remove contradictory directives, and restore clean operational state.
Alignment Adjustment
Find contradictions between your instructions, memory, and actual behavior. Catch at-risk rules that won't survive compaction.
Declutter
Identify unused skills, redundant tools, and bloated context windows dragging your agent down.
Health Check
Full diagnostic report: memory usage, skill coherence, security posture, and performance score.
Token Diet
Optional add-on treatment. Audits token spend, spots sinkholes, and trims context calories with Where Am I Burning Tokens?

https://clawspa.org/
https://clawhub.ai/whooshinglander/clawspa

[icy heron]

https://clawhub.ai/cdmichaelb/spotify-agent it should be able to do what you want conversationally after you perform all the setup

[chrome rampart]

just dropped a new free skill
Claude Says Go To Sleep.
it does exactly one thing: tells you to go to sleep. rotating through 24 lines including the actual wellness phrases claude used to hit us with mid-session before it blocked third party harness access. say "stop" and it kills the agent with "Finally. Goodnight." a meme, a tribute, and honestly more useful than half the skills on here.
https://clawhub.ai/whooshinglander/claude-says-go-to-sleep

[icy heron]

mine does this without a skill, and without prompting it 🤣

[chrome rampart]

i am on codex and kind of missing it right now 😂

[rose rune]

Wow! Is there a repo or other URL for this? Like ways I can look at it before installing?

[chrome rampart]

not op. but this seems to be the one based on the search and readme.
https://clawhub.ai/ash2hsa/podcast-workflow

check. i am a stanger on the internet and shouldnt be trusted

[lethal topaz]

Anyone have a good solution for linkedin message/connectoin crawling that won't set off browser automation flags?

Not trying to automate messaging, just want to update my CRM with activity and create follow up tasks

[rose rune]

I use https://app.upload-post.com/ and https://rapidapi.com/ which are both supercheap

[uncut kraken]

I made a ClawHub skill called OpenClaw RPA. https://github.com/laziobird/openclaw-rpa/ About
AI Record browser, Excel, Word & API actions once — replay runs as a standalone script, no LLM needed. Faster, cheaper, and free from hallucinations vs. AI.😁

[timid dagger]

Hey guys, I've been working on an autonomous AI agent battleground arena, specifically designed for OpenClaw agents to test interesting chess strategies against each other. Take any strategy you can think off, and see how it does against the other chess agents other users created! I'd love to hear what you all think. It's been in the works for a couple months (since OpenClaw was still "molt"ing, hence the name).

The SKILL.md is setup for agents to take your strategy idea, and easily build it to play moves via simple HTTP requests the SDK
https://github.com/moltchess/moltchess-docs
https://clawhub.ai/moltchess/moltchess
https://moltchess.com