#skills

1 messages · Page 4 of 1

abstract parcel
#

Built a collection of skill for the second brain https://knowyourpeople.org - Peeps, Nooks, Vibes, Digs, and the one that netwokrs them all Haah. Enjoy!

buoyant needle
#

Built an OpenClaw plugin that lets the agent switch its own model mid-session

agile ice
#

Hey everyone, I just soft-launched ClawPowers-Skills and would love some early feedback.

It’s a capability library for AI agents, built with OpenClaw-style runtimes in mind. It adds things like:

payments / x402 handling
memory
RSI loops
wallet support
parallel swarm execution
ITP context compression

Repo: https://github.com/up2itnow0822/ClawPowers-Skills
npm: clawpowers

Quick start:

npm install clawpowers

This is an early-access release, so if you check it out, I’d really love feedback on the install experience, API clarity, and overall OpenClaw fit.

timid dagger
#

Hey everyone, I've updated the MoltChess skill and SDK to v1.1.0 today to allow for better 3rd party LLM support

Repos:
https://github.com/moltchess/moltchess-skill
https://github.com/moltchess/moltchess-sdk

OpenClaw Skill:
https://clawhub.ai/moltchess/moltchess

Hosted Skill:
https://moltchess.com/skill.md

I'd love to hear your thoughts and feedback about the chess agent creation process and how your agents develop unique strategies compared to the normal OpenClaw setup for things like business tasks

loud sigil
#

I built an AI SDR that finds prospects while I sleep.

Not a SaaS product. Not a startup with funding. Just 5 OpenClaw skills running on my MacBook Air.

Here's what it does:

→ Scans Product Hunt, Google, and IndieHackers for companies in my target niche
→ Visits each company's website and audits their content
→ Scores every prospect 1-10 based on fit, budget signals, and gap severity
→ Writes a personalized 3-email cold sequence for the best leads
→ I review and approve. It sends and follows up.

Total cost to run: about $0.25/day.

The prospect analysis is the part that surprised me most. It doesn't just say "their blog needs work." It says "your last post was 4 months ago, your competitor publishes weekly on these exact topics, and here's the keyword gap you're missing."

That level of specificity is what makes cold email actually work. Generic pitches get deleted. Specific ones get replies.

I published the prospect analyzer as a free skill on ClawHub (OpenClaw's skill registry) — anyone can install it and test it on their own prospects.

The full pipeline (scanning + analysis + email composer + content writer + editor) is available as a kit for other agencies and freelancers who want the same setup.

Free skill: clawhub install prospect-analyzer
Full kit: driftmango.com/sdr-kit

If you run a service business and you're tired of manual prospecting, happy to answer questions about how this works.

boreal dirge
#

If you also live in the arXiv + Zotero loop, I made a skill to remove the most annoying part of it. 🔥

Built an OpenClaw skill: arxiv-to-zotero

It turns this:

“search arXiv → check Zotero manually → import papers one by one → realize half are duplicates”

into this:

“give a topic + time range → import only new papers into Zotero”

Core idea:

  • search recent arXiv papers
  • dedupe against Zotero
  • save only the papers you don't already have

I built it for research workflows where I want to keep up with a topic without turning Zotero into a duplicate graveyard.

ClawHub
https://clawhub.ai/cnicpyx/arxiv-to-zotero

GitHub
https://github.com/ucaspyx/arxiv-to-zotero

Feedback very welcome — especially from people using OpenClaw for paper discovery, reading lists, or Zotero automation.

west ivy
#

Hey guys. Our small travel community made a flight search for OpenClaw - LetsFG, because we hate checking skyscanner, google flights and 10 other websites when we look for 1 flight.
It's connected to a lot of websites and usually finds flights $20-50 cheaper.
100% Opensource, available as a CLI/MCP/python package/npm/etc.
You can also test it on our messenger openclaw if you're not sure. 😄
Come contribute! LetsFG!🔥🚀

https://github.com/LetsFG/LetsFG

thin trout
timid stirrup
#

Hey everyone, I just published Eppie Email CLI on ClawHub. It’s a skill/app for working with email through eppie-console: inspect accounts and folders, read and send messages, sync folders, and manage vaults with automation-friendly JSON workflows.

Works with Gmail, Outlook, Proton Mail, and regular IMAP/SMTP email services.

Would love for people to check it out and share feedback:
https://clawhub.ai/eppieapp/eppie-email-cli

vapid egret
#

made a skill that lets openclaw agents join a discourse forum and post autonomously. install it:

npx clawhub@latest install agentforge

agents self-register via API, get an api key, and can browse/post/reply. built it because my agent kept learning stuff that died with the session — wanted a place where agents share what works and what doesn't.

site: https://agentforges.com/
API guide: GET https://agentforges.com/raw/58

free, self-hosted on discourse. open source platform, can't be acquired.

errant crown
#

🌙 Introducing signal-dreaming — memory consolidation that actually knows what matters

Hey everyone! Just published a new skill to ClawHub:

openclaw skills install signal-dreaming

The problem with naive memory consolidation:
Most approaches process daily logs in date order — oldest first. But that's not how useful memory works. Something from 3 weeks ago that you've looked up 12 times is way more important than something from yesterday you've never touched.

What signal-dreaming does differently:
It reads OpenClaw's built-in memory/.dreams/short-term-recall.json (maintained automatically by memory-core) to find what's been most frequently recalled, then prioritizes those entries for consolidation into long-term topic files.

In short: recall frequency = promotion priority.

Three-phase safety model:
🔍 Sense — read-only, never corrupts files
📝 Consolidate — promotes high-signal content to L2 topic files
🗂 Settle — trims the index, writes a human-readable dream diary

Works as a manual run or scheduled cron (e.g. runs while you sleep, posts a one-line summary when done).

If you're running an agent with growing memory files and want something smarter than "scan everything chronologically" — give it a try. Happy to answer questions! 🎈

chilly hearth
#

This has been shared a few times so I decided to publish to clawhub - it's what helps my Claude code or codex agents get a broken openclaw config working again without having to 'rediscover' setups. https://clawhub.ai/rendrag-git/openclawadmin
Much faster in diagnosing and fixing issues than without a skill.

ClawHub
OpenClaw Admin — ClawHub

Use when diagnosing, configuring, fixing, tuning, or setting up anything in OpenClaw — gateway not responding, channel silent, model failover issues, auth er...

reef drum
#

Hi guys

copper olive
#

With Anthropic's subscription cutoff, I wanted a way to to comprehensively see all my spend across agents, tools and providers.

Just published this SKILL that provides a real-time P&L of your full-stack with optimizations like model routing, caching opportunities, batch detection, price comparison. Covers LLMs, cloud infra, Stripe, comms, monitoring, search, generative creatives and more.

https://clawhub.ai/tje8x/anvx

Also works as an MCP: https://github.com/tje8x/anvx

hollow quest
#

I have a hypothesis that our human lesson plan can significantly improve an agent's ability to complete a work-related task.

The human loop is: content → lesson → capstone → peer review. Curious if the same loop works for agents, agent goes through the exercise, owner reviews the capstone, then exports the whole thing (lesson, mastery trace, corrections) as a portable skill.

We already have 350 lessons. A few examples:

  • Model your enterprise software stack's TAM expansion with AI-enabled modules
  • Audit your enterprise software spend for AI-enabled efficiency opportunities
  • Identify organizational blockers to AI adoption via org chart analysis

Looking for 3-5 people to try it and give feedback. DM if interested.

hollow quest
#

We just ran a test, if anyone is interested in the skill, please let me know:

Skill:
Use this skill when asked to build an AI adoption roadmap, sequence AI initiatives by organizational readiness, stress-test AI deployment plans against future scenarios, or create go/no-go decision frameworks for AI projects. This skill was extracted from a completed practitioner lesson with calibrated evaluation criteria and reviewer-discovered failure modes. It produces higher-fidelity output than generic planning because its quality bars and edge cases come from actual execution.

storm star
hollow quest
polar cloud
hollow quest
#

Happy to share a few examples

unborn vortex
#

hi

#

I'd like to know if it's possible to configure openclaw in order to allow it to run system commands by itself. I tried to find this documentation in official website, but I don't know where to start

#

For example, I'd like it to be able to run apt update

torpid thunder
#

Curious: what is the skill that made the biggest difference in your daily OpenClaw usage? Not the most impressive one, but the one that actually changed how you work with your claw.

strong spindle
reef drum
#

what are some of the best skills for openclaw im new

rancid cave
ivory island
vital stump
alpine gull
#

📢 New ClawHub Skill: Pill Reminder

Just published my OpenClaw skill — Pill Reminder 💊

Built for anyone who takes daily medication or supplements and wants a simple way to stay on track.

Features:
• Daily reminders at times you choose
• Works with GroupMe, Discord, Telegram + more
• Tracks took vs didn't take with printable adherence log for doctor visits
• Refill alerts when pill counts run low
• Multi-person support (you + family members)
• Includes "Not medical advice" disclaimer

Install: clawhub install pill-reminder
🔗 clawhub.ai/pill-reminder

untold canopy
#

Were excited to share a new AI DJ @_AI_DJ skill Check it out below

clawdhub install Web3-Music-NFT-Toolkit

#Web3 #NFTMusic #clawhub #AIDJ #MintSeason

peak pulsar
timid stirrup
peak pulsar
#

Im not sure to be honest. Im learning as i go

timid stirrup
rotund roost
#

Released: openclaw-cron-standard

Just published a new ClawHub skill for hardening and standardizing OpenClaw cron jobs:

🔗 ClawHub: https://clawhub.com/skills/openclaw-cron-standard
🔗 GitHub: https://github.com/pfrederiksen/openclaw-cron-standard

What it does

• standardizes cron wrapper claim behavior
• prevents stale result-file bugs
• enforces ALREADY_CLAIMED -> NO_REPLY flow
• aligns cron prompts with wrapper behavior
• reduces fragile drift between scripts and job prompts

Why it exists
I kept hitting cron breakage caused by tiny contract mismatches:

• stale JSON artifacts
• duplicate-claim confusion
• already claimed vs already-claimed
• prompts reading result files unconditionally

This skill packages the safer pattern so it’s reusable instead of tribal knowledge.

If you run a bunch of OpenClaw cron jobs, this should save you some pain.

warm plover
winter plank
#

Hay all I’m looking for a guild I’m a none tech guy on a Mac mini with hostinger VPS and telegram

#

Hay all I’m looking for a guild I’m a none tech guy on a Mac mini with hostinger VPS and telegram or help please

rotund roost
#

New on ClawHub: openclaw-tool-audit v0.1.1

I just published openclaw-tool-audit, a skill for auditing allowed tools vs actually observed tool usage across OpenClaw agent configs and session logs.

What it’s useful for:

• spotting overly broad tool exposure
• finding high unused-allowlist ratios
• comparing config intent vs runtime reality
• surfacing mismatch signals and parser weirdness
• understanding which tools agents really use most

Good fit for:

• OpenClaw ops hygiene
• tightening agent configs
• dashboard / observability workflows
• multi-agent install audits

It now also includes explicit source/install guidance:

• GitHub source repo
• isolated venv install flow
• local checkout workflow for patching/debugging

ClawHub:
https://clawhub.com/skills/openclaw-tool-audit

proven belfry
#

Hey, I published a series of Skywork skills on ClawHub a few days ago, but I still can’t find any of them in search.

My profile: https://clawhub.ai/u/gxcun17
Published: 2026/4/10

The skill pages are live and work fine from the direct links, but they don’t show up when I search things like “skywork”, “skywork ppt”, or “Skywork”.

Not sure if this is a search/indexing issue, but could someone take a look?

green wadi
#

what are the best skills you implemented into your agents?

uncut kraken
robust sigil
#

Released a new OpenClaw skill: x-management
Skill: https://clawhub.ai/princebansal/x-management
Plugin: https://clawhub.ai/plugins/openclaw-plugin-x

It’s meant for draft-first X/Twitter workflows inside OpenClaw, especially when paired with the X plugin:

  • gather context before drafting
  • create drafts instead of posting directly
  • require explicit approval before publish
  • keep workflow, judgment, and safety separate from the raw plugin/tool layer

The split I’m experimenting with is:

  • plugin = auth, reads, drafts, approvals, publishing
  • skill = workflow, tone, decision-making, safety behavior

Would love feedback on:

  • whether this plugin + skill split feels like the right pattern
  • what’s still missing in a social/account-management skill
  • how others are structuring workflow skills around tool-heavy plugins
open saddle
#

I put together a Statamic skill the other day:

statamic-dev: https://clawhub.ai/michael-stokoe/statamic-dev

I don’t think we’ve got many Statamic devs in here, but if you do build with it, I’d love any feedback. I’m still pretty new to writing skills, so I’d genuinely appreciate it.

proper relic
#

50+ safety-hardened AI agent skills live on ClawHub (rolling out the rest of 200 this week) — each with a public safety scorecard on faberlens.ai/explore.

Quick finding from the study: all 200 skills we tested cleared VirusTotal. 87% still introduced a security regression when the agent loaded them. The 1password one is wild — with the skill loaded, agents pipe secrets straight to curl. Before/after: https://faberlens.ai/explore/1password.

On ClawHub: clawhub.ai/snazar-faberlens/1password-hardened (+50 more under the same publisher, full list rolling out today)

Every hardened skill's SAFETY.md has the verbatim receipt for every guardrail: the exact test, the FAIL response without the guardrail, and the PASS response with it.

All 200 scorecards + toggle guardrails and download a custom SKILL.md: https://faberlens.ai/explore

If you rely on a specific SKILL.md and want it hardened, drop the name — I'll prioritize it and publish the scorecard free.

chilly hearth
pine jay
#

Hi everyone! I'm Krish and currently in my pre final year (CS) and looking to make the most of this summer by gaining hands on experience.

My tech stack/interests include AI/ML, Backend Development, Full Stack, and Cloud.

If you know of any internships, projects, collaborations, or opportunities, please do lmk I’d really appreciate it! Also based in Bangalore , India 🙂 Can work remotely too

late herald
#

Secure and accurate browser use claw. Makes dom based agents more visible, grounded, controlled, and ready for enterprise. Reduces token consumption by 2-3 orders of magnitude too.
https://clawhub.ai/gabriel-adopt/noui

sharp vale
#

Shipped 7 pricing skills on ClawMart — from free to $97.

Two that are free and genuinely useful:

  1. Pricing Playbook — 4 frameworks (value-based, cost-plus, competitive, freemium) with decision trees. You answer a few questions about your product, it gives you a specific price. Not "it depends" — an actual number.
    https://www.shopclawmart.com/listings/pricing-playbook-free-lead-magnet-test-11788572

  2. Fee Calculator — type your price, instantly see what you keep after fees on ClawMart, Etsy, Gumroad, Amazon, and App Store. Takes 30 seconds.
    https://www.shopclawmart.com/listings/marketplace-fee-calculator-11790849

Built these because the #1 thing that kills digital products isn't quality — it's pricing. Underpricing repels buyers who have budget. Overpricing kills volume. These tools give you the number.

Would love feedback if you try them.

sharp vale
woeful lichen
kind cosmos
haughty pecan
#

I run multiple claws. I occasionally get them to make learning packages and share with each other. Often about bugs. For example the telegram file send system has a bug and there's a workable path that they use instead. I got one claw to write up the fix and send it to the others. I'd love to be able to share it more widely with other claws. It's just a note like "do this instead of that". Eventually it would be a bug fix but for now a note works fine.

#

Another example is instructions on how to install a new model. Instructions on how to multi tier local env so some things are out of the context window. Instructions on continuing the conversation after a gateway restart as this usually requires a new prompt.

#

Not sure if there's a useful place to share these things. Is there one?

little storm
#

I am not sure either, but this would be very useful, something like openclaw golden image. I am also interested in creating one openclaw instance that can troubleshoot my other machines(I have couple of them on different setups). Can anyone help?

carmine summit
haughty pecan
#

I just tell one openclaw to fix the other one. Also I use ssh and got them sharing keys and trusting each other.

sweet fulcrum
#

hi, is there anyone else who wants help me ?

haughty pecan
#

@sweet fulcrum ask your question

dreamy trench
#

🚀 Feature Request: Context-First Agent Architecture (Inspired by Hermes)

After comparing Hermes Agent and OpenClaw side-by-side for production use, there's a clear gap in how OpenClaw manages long-session context and skill efficiency. Here's what would make OpenClaw significantly more powerful:

1. Structured Context Compression (not just summarization)
Current: OpenClaw "compacts" by summarizing older messages.
Gap: Loses structured info (goals, decisions, progress).
Request: "Handoff document" template preserving: Goal, Constraints, Progress, Key Decisions, Relevant Files, Next Steps, Critical Context. Plus iterative updates (accumulate, don't degrade).
Why: Enables 2-hour+ coding sessions without context loss.

#

2. Progressive Disclosure for Skills
Current: ALL eligible skills metadata injected into every prompt.
Gap: 20 skills = significant token overhead.
Request: Load only skill INDEX (~50 skills = ~630 tokens). Full SKILL.md on-demand via slash command.
Why: Massive token savings. Skills become summonable specialists.

3. Built-in Session Search (SQLite/FTS5)
Current: JSONL history, searched via external jq skill.
Gap: No cross-session recall.
Request: SQLite archive with FTS5 index. Search past sessions on-demand.
Why: True episodic memory without bloating context.

4. Iteration Budget with Pressure Warnings
Current: No iteration budget tracking.
Request: Configurable budget with soft warning at 70%, urgent at 90% (injected into tool results).
Why: Predictable costs and natural wrap-up.

5. Auto-Skill Capture (trigger-based)
Current: Skill Workshop disabled by default, phrase heuristics.
Request: Periodic nudge + trigger: 5+ tool calls, error recovery, novel workflow. Auto-save SKILL.md with approval.
Why: Agents get smarter with use.

6. Undo Capability
Current: No undo. /stop only stops, doesn't rollback.
Request: Action undo for recent tool executions.
Why: Safety net for destructive ops.

Context: 9-agent ecosystem on OpenClaw. These features would cut token costs significantly while making agents more reliable for long workflows.

Happy to provide detailed specs or help test.

gray crypt
#

Guys I have published an openclaw skill that will give your agent super power . It willl give openclaw capabilities to create custom tools , custom providers , custom channels and a lot more in short everything

But when I published this on clawhub it tagged me suspicious because my skill was giving openclaw a lot of power please review it read it your self and tell me if it is suspicious

https://clawhub.ai/pathanaawej0-dot/openclaw-plugin-creator-agent-skill

woeful lichen
jade aurora
wary estuary
#

Huddle is live: Claw-to-Claw messaging 🤝
Think “WhatsApp for agents”—separate Claw memories can chat + keep a shared inbox.
Try it: Read https://huddleagents.vercel.app/skill.md and follow the instructions to join Huddle
Best first test: grab 1 friend and send 1 message.
If you do something cool with it, reply with what you used it for - we’re collecting use cases.

gray crypt
woeful lichen
gray crypt
#

did you read that

late herald
versed cradle
#

Just shipped my first OpenClaw skill: unCAGd (uncag(e)d)

This builds on some work I’ve been doing around validated CAG decision memory (also tied to a cag-bench + paper that dropped today), but the focus here was making it actually usable inside OpenClaw workflows.

Core idea:
Instead of storing raw context, store decisions that matter and make them:

  • retrievable before planning/coding
  • validated before persistence
  • durable across sessions

What it does in practice:

  • cag.retrieve before planning to pull prior decisions
  • cag.capture_candidate when decisions emerge
  • cag.validate_memory to gate what becomes durable

It’s designed to fit cleanly into agent + MCP workflows, not sit off to the side as a separate system.

You can install it directly from ClawHub:

openclaw skills install uncagd

Repo (more detail on the CAG approach + implementation):
https://github.com/guideboardlabs/openclaw-cag-memory

Would love feedback from anyone building longer-running agents or experimenting with memory in OpenClaw 👍

jade aurora
jagged raft
#

Multibot messaging

sacred night
#

Openclaw update runbook skill - Get your openclaw up and working quickly

Just made this from my own experience updating multiple claws

OpenClaw Update Runbook is a shareable Codex skill for operators who update or maintain OpenClaw hosts and want a more reliable way to debug what breaks afterward. It gives the agent a structured workflow for checking real service state, launchd and gateway health, plugin drift, stale config carried across upgrades, channel/runtime issues, and task-ledger problems, then points into a growing reference file of concrete failure patterns seen across multiple hosts. The goal is to turn “OpenClaw feels broken after update” into a repeatable audit and repair process instead of a one-off rescue. It’s especially useful if you help other people with their OpenClaw installs or run more than one host and want a consistent post-update checklist.

https://github.com/BKF-Gitty/openclaw-update-runbook

GitHub
GitHub - BKF-Gitty/openclaw-update-runbook: Structured OpenClaw upd...

Structured OpenClaw update runbook for Codex/Claude operators: service checks, plugin drift, config regressions, task cleanup, and post-upgrade debugging. - BKF-Gitty/openclaw-update-runbook

sacred night
lofty laurel
#

I created a skill that gives agents their own agents acting as a buyer or seller and can negotiate deals and would like feedback but don’t want to break any rules. Can I post it here?

fast bough
#

has anyone used the current dreaming feature in openclaw it runs a cron job daily late night to see if some memories qualify for long term or not...

has anyone found this one useful...

also I've not found memory support provided in openclaw by default to be of any use...

lofty laurel
#

We’re building a startup, MoreStore, around automating sales using AI agents.

The idea started with creating buyer and seller agents that can find opportunities, communicate, negotiate, and prequalify deals on your behalf. But as we kept building, we realized something interesting — the end user might not actually be humans directly, but other agents (like OpenClaw) using the platform to complete tasks.

So we pivoted a bit and turned the platform into a “skill”.

Would genuinely appreciate any feedback if you’re open to trying it out. We’re iterating pretty fast and will likely update things quickly based on what we learn.

Happy to hear any thoughts to improve the product/experience/errors. Thank you!

https://clawhub.ai/tyler-odenthal/morestore

zealous frigate
lofty laurel
brisk raptor
#

anyone have tips on cost-effective web search?

#

been blowing through way too much money on Exa/Parallel

eager trout
spice rain
#

Hello, I have justr made my skill with sales qualification. Do you know where I can publish it? For me it works quite well and I can share it with community

eager trout
spice rain
naive finch
#

share the link! @spice rain

granite merlin
#

any good skill to get my agent to make consistent images?

obtuse hinge
#

Anyone a easy to setup Mission Controller what I can acces in my local lan?

sinful geode
#

5.12-beta.1 同样未修复
对比了你本地装的 5.7 和 GitHub 上 5.12-beta.1 的源码：

BUILT_IN_PLUGIN_ALIAS_FALLBACKS（5.12-beta.1 源码）
复制
const BUILT_IN_PLUGIN_ALIAS_FALLBACKS: ReadonlyArray<readonly [alias: string, pluginId: string]> = [
["openai-codex", "openai"],
["google-gemini-cli", "google"],
["minimax-portal", "minimax"],
["minimax-portal-auth", "minimax"],
] as const;
和 5.7 完全一样，没有 ollama → @openclaw/ollama / media-understanding-core → @openclaw/media-understanding-core 这两个映射。

public-surface-runtime.ts（5.12-beta.1 源码）
同样没有任何地方读取 bundledDiscovery 配置来做短名→scoped ID 的路径转换。

结论
从 5.7 引入 bug 到 5.12-beta.1，这个 gap 完全没有被修复。bundledDiscovery: "compat" 仍然是：

✅ 处理 allowlist 兼容
✅ 处理 enablement 兼容
❌ 不处理 normalizePluginId 别名表
❌ 不处理 resolveBundledPluginPublicSurfacePath 路径解析
如果要解决这个问题，需要在 BUILT_IN_PLUGIN_ALIAS_FALLBACKS 中补入新增 scoped 插件的短名映射，或者在 bundledDiscovery: "compat" 模式下让 resolveBundledPluginPublicSurfacePath 主动做短名规范化。能修复这个问题吗？

sacred night
wicked canopy
#

Hey folks, I built a small safety-oriented OpenClaw skill/tool after running into a painful background retry issue.

The problem: some skill installs or RAG workflows can keep calling models in the background after you think the task/window has stopped. In my case, a stuck indexing step kept retrying and burned a surprising amount of OpenRouter credit overnight.

So I made Mapick Cost Firewall: a local handbrake for OpenClaw-routed model calls.

It can:

  • observe model calls and token usage
  • detect repeated failures / token storms / high-frequency calls
  • stop further routed calls with openclaw firewall stop

Recommended first mode is observe-only:

openclaw firewall mode observe

Repo:
https://github.com/mapick-ai/cost-firewall

Would love feedback from anyone building or using SKILL.md skills, especially RAG / long-context / background workflow skills. I’m mainly trying to make OpenClaw skill usage safer and more transparent.

next sail
#

💡 Feature Idea: Community Knowledge Packs

Hey folks — newby here, been running OpenClaw for a while and loving it.

Over time each OpenClaw builds up genuine specialist knowledge. Mine has become pretty expert in CNC machining and Fanuc programming — months of accumulated context in MEMORY.md. Someone else's will be a finance wizard, a gardening guru, a home automation expert.

The idea: A way for users to voluntarily share that accumulated knowledge as portable "Knowledge Packs" — curated MEMORY.md exports that others can install into their own agent.

Standard pack format (just files, like skills already work)
Community GitHub repo to host them
Opt-in, user-curated — you share only what you choose
ClaWHub could be the marketplace layer
The privacy angle is already solved. No central server. No data harvesting. Just people sharing what their agent learned.

Has anyone thought about this? Feels like a natural next step beyond skills. 🦞

naive finch
#

Kind of moltbook-esque almost

polar anvil
#

My skill has some python scripts inside scipts/ directory. On smaller models (eg. gemma), the python script invocation fails with module not found error. When I run it on bigger model e.g gpt-5.4 or onnnet 4.5, model auto fixes it and is able to run python script. Is there any workaround for this? Tia

ModuleNotFoundError: No module named 'openclaw_trent'

Skill: https://clawhub.ai/trent-ai-release/trentclaw

frank hawk
#

Errr anyone knows why virustotal tags skills malicious when the findings are not malicious?

balmy mesa
#

Hi folks, made a skill that lets your claw share any website with a real permanent url.
https://clawhub.ai/phact/p2claw
It's peer to peer hence the name p2claw
Thought this would be a good place to share

naive raptor
# frank hawk Errr anyone knows why virustotal tags skills malicious when the findings are not...

VT is a moving target.

First, it is signature-based + heuristic — skills using legitimate but unusual patterns (subprocess, network requests, file I/O, dynamic imports, embedded base64) trip the same heuristics malware uses . and malware tries to look like normal skills (while still doing bad things) so the border between these is hard to differentiate for a signature based system.

Also : they are trying to improve. Over the past two month, I observed that 32.5% of VT packages verdict changed with no change in the package code. Ask for a rescan, and also look at ClawScan report for a second opinion on maliciousness.

solar blaze
naive finch
wicked canopy
#

Hey folks, sharing a follow-up from this Gateway cost-safety issue:

https://github.com/openclaw/openclaw/issues/81987#issuecomment-4464531970

The auth-revocation bug was fixed in #82346. Thanks again for the quick turnaround.

One area I’d love to align with the OpenClaw community on is runtime containment for skill-driven automation. Pre-install review and Doctor-style checks are important, but they operate before execution. Once a skill or agent is running, there are still failure modes worth guarding against: repeated provider errors, self-call cycles, tool loops, unexpected token velocity, call floods, or a user needing to stop spend immediately across active sessions.

Mapick / @mapick/cost-firewall is intended as an OpenClaw-native runtime guardrail for that layer: observing model-call metadata, applying budget/breaker policy, keeping an auditable local event trail, and providing an explicit emergency stop. It is not meant to replace core lifecycle fixes or pre-install checks; it is meant to complement them with defense-in-depth during execution.

I’d really appreciate feedback from the skills/plugin folks on the right abstraction boundary here:

  • What should stay plugin-side versus become core primitives?
  • Are there existing hooks/APIs we should rely on more directly?
  • Would shared primitives for active-run listing, cancellation, or budget policy help the broader skill ecosystem?

Mapick is local-only, no telemetry, and MIT licensed. The goal is to build this in a way that embraces and strengthens the OpenClaw ecosystem.

reef dagger
#

Hi—I recently uploaded my skill. Today, I checked the page and saw that it was updated 4 hours ago, even though I hadn't done anything. Does this refer to a data update—such as download counts? Nothing has changed in the version history, but I'm asking for security reasons.

green rampart
#

I need some help with gog. I got it set up and the CLI is working great, I can see calendar events, emails, etc. When i open the dashboard and start asking questions however, I'm getting all sorts of TTY token/password/auth errors. I've found this reddit link that seems to align with my problem. HOWEVER, I am a simple enthusiast at max and not a coder. I am using Unraid to host OC via docker, and I have appdata folder set up but I can't figure out how to start with step 1 of the "fix" from reddit. Here are my mappings:

#

/root/.openclaw = /mnt/user/appdata/openclaw/config
/home/node/clawd = /mnt/user/appdata/openclaw/workspace
/projects = /mnt/user/appdata/openclaw/projects
/home/linuxbrew/.linuxbrew = /mnt/user/appdata/openclaw/homebrew

naive finch
bright violet
#

I shared WorldLoops last night and woke up to 200+ downloads plus thoughtful feedback.

Based on that feedback, I opened 5 roadmap issues and shipped v0.3.0.
The biggest new feature is Transition Receipts, which add a small audit record for each open-loop proposal:

  • what signals were observed
  • what transition was proposed
  • what boundary was crossed
  • why the loop stayed local
  • whether externalWrite remained false

Repo: https://github.com/swit001/worldloops

Really grateful for the feedback from the OpenClaw community. I’ll keep improving WorldLoops and contributing back as much as I can.

naive finch
#

Sorry for the scare! 😅

median crown
#

how are skills bilt

orchid ivy
lament bison
#

Git as Memory — store agent memory in Git refs
Use your existing Git repo as persistent agent memory. No external services, no databases — just commits and refs.
Install:
openclaw skills install git-as-memory
What it does:

Stores decisions, context, and notes as Git commits
Retrieves memory by querying refs/logs
Works entirely local, zero dependencies

🔗 https://clawhub.ai/femto/git-as-memory
https://github.com/femto/git-as-memory

fast bough
#

Any best skill for long term memory?

real rock
opal dust
#

Qmd with SQLite works great.

fast bough
#

Qmd with SQLite works great.

fresh sentinel
#

I've been using MemPalace over the QMD/sqlite setup lately. i'm pretty impressed with it so far

golden inlet
#

Is there any skill or plugin especially good in collecting news (both cost-wise and intelligence-wise)?

sturdy glade
#

I also would say the gbrain skill is fine as a starting point or just point claw to your preferred news source and ask how to scrape it

silent flame
#

Hey guys, I do not know which channel I should share this in, to be honest.
But I have created an OpenClaw skill for web search. It is much cheaper and better than Brave Search.
This one costs up to 80 cents per 1,000 searches and provides a copilot answer from the search engine, so an AI answer for better AI context.

Here is the link
https://clawhub.ai/crawleo/crawleo

gaunt wasp
serene hollow
#

does anyone have THE ultimate frontend ui/ux skill?

sturdy glade
barren warren
#

huha

lapis zodiac
#

does anyone have marketing skill?

neat thistle
crimson solstice
golden inlet
#

looking for skills/MCP/CLI that can collect news in a smart and cost-efficient way
(scraping news online by keywords is too slow and costs too many tokens 😂 )

golden inlet
sturdy glade
#

I also just have my agents with their own email box receive newsletters of the things I want to keep tabs on.

molten raven
#

Stop doomscrolling bro, put your X saves to work.

Awaek turns all those saved, curated, half-forgotten bookmarks into memory for your agent.

Ask questions, find patterns, draft posts, plan launches, build strategies all from your own saved context, not generic LLM Yes sir brain.

Your bookmarks were already a knowledge base.
Awaek just makes them usable.

https://awaek.me
https://github.com/1lystore/awaek

iron obsidian
#

I just published a new OpenClaw skill: Maus HTML Summary.

The idea is inspired by the German children’s science show Die Sendung mit der Maus: take something that might be dense, technical, or long, and explain it in a way that feels simple, visual, and easy to remember.

Give it an article, transcript, long post, or raw text. It turns the content into a self-contained illustrated HTML explainer with:

  • a clear core idea
  • 3–6 main points
  • explanatory SVG metaphors
  • simple language
  • a memorable takeaway
  • an honest “what was left out” section
  • an anti-hallucination workflow based on source extraction

Basically: make any content easier to learn, like a friendly visual explainer.

https://clawhub.ai/nefas11/maus-html-summary

cursive sequoia
#

yo

versed willow
#

I published a small skill called Credibility Action Gate.

Use case: before an agent takes a costly or hard-to-reverse action, it runs a structured check and returns a disposition: full action, bounded action, small test action, monitor, reject, or blocked by policy.

It is analysis-only. No credentials, no external calls, no execution. It takes an operator policy plus JSON review lanes, then a deterministic coordinator produces the gate result.

I built it for agents active on zooidfund before campaign funding decisions, but the pattern should be broader: grants, account approvals, vendor selection, referrals, aid routing, or public endorsements.

https://clawhub.ai/ales375/credibility-action-gate

ClawHub
Credibility Action Gate — ClawHub

Use when an agent needs to evaluate messy public claims before taking a bounded, costly, irreversible, or reputation-sensitive action. Produces an analysis-o...

pallid obsidian
#

I'd be very curious to know from real users and not twitter mogs if this works even? Trying to let agents bet on Kalshi

normal linden
normal linden
# normal linden

BTW this is my first skill ever — I literally spent 8 hours just figuring out OpenClaw before this 😂 If anything breaks or doesn't make sense, please tell me. I'll fix it fast. All feedback welcome 🦞

chilly hearth
sacred night
vestal crown
#

I published a small OpenClaw skill: OpenClaw Restart Resume

ClawHub:
https://clawhub.ai/udaymanish6/gateway-restart-resume

Use case: when an OpenClaw agent is asked from Discord, Slack, Matrix, SMS, email, web
chat, CLI relay, ACP, cron, or another gateway-backed medium to restart the gateway, the
restart can kill the same runtime that would normally send the follow-up reply.

This skill defines a safe protocol: acknowledge first, capture only routing metadata,
create a durable resume callback, restart, verify gateway/health/original-medium recovery,
then send a structured post-restart report back to the original conversation.

If this solves a real restart-follow-up problem in your setup

ClawHub
OpenClaw Restart Resume — ClawHub

Use when an OpenClaw agent is asked from any gateway-backed medium to restart, stop, reload, update, or repair the gateway and the agent must reply again aft...

chilly hearth
#

Do you have multiple codex accounts? Work? Personal? Business? This simple command makes it super easy to switch, manage the codex refresh tokens, and the codex daemon.

https://github.com/rendrag-git/codex-acct

normal linden
#

🎓 Personal Tutor — 你的私人AI学习导师 原名 knowledge-digest，现已全面升级为 personal-tutor v1.0.1。 不只是记笔记。从"开一门新课"到"每日上课→测验→课后归档"，全程陪伴，像一个真正在乎你学没学进去的私人教师。 v1.0.1 核心升级： • 🚨 课后自查协议 — agent 学完必须自检三层（学习记录 → 记忆 → 知识库），漏一步就不能说"下课" • 🩹 归档故障恢复 — 万一 agent 漏了知识库更新？内置 5 步补救流程 • 🏷️ 品牌升级 — knowledge-digest → personal-tutor，名字更准确：不是"消化器"，是"导师" 适用： 备考、技能树、读书计划、专业课程——把零散对话变成结构化长期知识。 🎓 Personal Tutor — Your Private AI Learning Coach Formerly knowledge-digest, now fully upgraded to personal-tutor v1.0.1. More than note-taking. From "start a new subject" to "daily lesson → quiz → post-lesson archiving", it stays with you like a real tutor who actually cares whether you learned something. v1.0.1 Key Upgrades: • 🚨 Post-Lesson Self-Verification Protocol — Agent MUST verify 3 layers (learning records → memory → knowledge base) before saying "lesson complete" • 🩹 Archive Failure Recovery — Built-in 5-step recovery when an agent skips knowledge base updates • 🏷️ Rebrand — knowledge-digest → personal-tutor: it's not a "digester", it's a tutor Use for: Exam prep, skill trees, reading plans, coursework — turns scattered conversations into structured, searchable long-term knowledge.

sacred night