clawtributors

[final ocean]

also can be skills etc

[noble moth]

Attached in Summary section. https://github.com/vezaynk/openclaw/pull/1 Still need to do some PR verification though

[modern kestrel]

watched the video, interesting feature. I didn't know telegram could do inline queries like that. read some / skimmed rest of the code and first feedback is: you better vibe up some tests homie 😂 i can't comment as to whether this feature matches the vision doc (I think it fits under - Improving support for major messaging channels (and adding a few high-demand ones) in Next Priorities). The 10 minute timer caught my eye and I'm still dizzy from it.

[lean seal]

@rugged light Hi just pinging with https://github.com/openclaw/openclaw/pull/27343 was hoping to get @ClawTributors role as per FAQ.

Also is this the right place to bump a feature request PR that fixes it.

https://github.com/openclaw/openclaw/pull/27812 in regards to this Feature request https://github.com/openclaw/openclaw/issues/9837

[foggy stream]

Opened 2 new PR's addressing open issues related to Gemini embedding.

PR #28361 - fixes the Gemini batch embedding infinite polling loop (related issue #15738).

• Reads metadata.state instead of status.state so async batch jobs actually complete instead of polling forever.

PR #28369 - adds retry with backoff to Gemini synchronous embedding calls (related issue #15738, #26069).

• Gemini was the only provider without retry — now 429s and 5xx errors get 3 attempts with exponential backoff instead of crashing immediately.

[foggy stream]

Fix for breaking change in .26 that broke replies in Mattermost: https://github.com/openclaw/openclaw/pull/28477

[open bluff]

i'm wondering if someone can figure out how to add this as a feature https://github.com/pyro0ownz/keyrotator/blob/main/key_rotator.py and https://github.com/pyro0ownz/Keymanager i noticed when you try to use more than one key the instance cries about api key limit hit but then when you use a single key it works whats the deal with that

i'm hoping i dont have to use claude code gemini cli and codex to tear apart the codebase and fix it manually

[past haven]

Hi folks! I have consistent CPU usage spikes every time claw starts. I profiled the problem and had codex fix it. The issue is mainly because the entire plugin-sdk is monolithically loaded every time. The PR splits this into lazily-loaded components, fixing the issue (confirmed with a local build)

Could any maintainers please look at it? Thank you! https://github.com/openclaw/openclaw/pull/28620

[sonic ermine]

folks posting their prs in here right after they were told not to by a maintainer : please dont do that

i'm not trying to minimod, but i understand why it's worth it to keep this channel open for discussions instead

give more than you take y'all 🦞

[shadow cloak]

I've made a bit of a monster sized change so before it even gets reviewed I thought I'd ask how much people would want this feature or if I should just maintain my own fork.

I've added a bwrap based backend for sandboxing as it allows unprivileged sandboxing with user namespaces quite easily (I'm running the gateway in a container with no permissions at all). Is anyone else interested in this use-case?

The changes are here: https://github.com/openclaw/openclaw/pull/28599 (including PR to provide context with the PR description, but if posting PR links is discouraged let me know)

[tawny vine]

Sorry I had missed this - deleted my post.

[crimson wren]

This PR has been open for about a week, so I wanted to follow up politely.

I’ve been using and maintaining this feature heavily in daily workflows, and I’ve done extensive manual testing across multiple agents. In practice, it significantly improves my Feishu doc workflow and overall user experience.

I’d really appreciate a merge when possible, so I don’t need to keep rebasing and adapting it against main repeatedly.

For current status: the only remaining CI issue appears to be unrelated infrastructure/runner instability (not a functional regression in this PR).

Thanks a lot for your time and for maintaining OpenClaw 🙏

https://github.com/openclaw/openclaw/pull/20304

[white galleon]

i can't find evidence for this in the current codebase but i'll keep looking; if you are able to profile the call stack to identify the source that would be great!

[keen spruce]

I can't find it in the code either, I only see the effects:

...
Feb 27 13:17:40 DietClaw snoopy[216894]: rp=openclaw-gatewa/185565 uid=0 tty=(none) cmdline=ip neigh show
Feb 27 13:17:55 DietClaw snoopy[216902]: rp=openclaw-gatewa/185565 uid=0 tty=(none) cmdline=ip neigh show
Feb 27 13:18:10 DietClaw snoopy[216910]: rp=openclaw-gatewa/185565 uid=0 tty=(none) cmdline=ip neigh show
...

I suspect it comes from one of the dependencies, but my node.js/Typescript knowledge is very spotty in that area. I'm a Perl/Java guy...
(And yes, uid=0. openclaw has a whole VM for itself here.)

[shadow cloak]

Is it potentially bonjour for the mDNS advertiser?

[white galleon]

can you try tracing it?
something like

readlink -f /proc/185565/exe
tr '\0' ' ' </proc/185565/cmdline; echo
pwdx 185565
systemctl cat openclaw-gateway 2>/dev/null || systemctl --user cat openclaw-gateway
strace -f -e execve -s 200 -p 185565

[white galleon]

on linux we use avahi-browse, not ip neigh show

it also repeats every 15s; i could only find code that uses either 30s or 60s in the codebase for discovery

which makes me think it's something else triggering that

[keen spruce]

I've injected a fake "ip" that outputs damamged data in hope of triggering an error, but no luck that way

[shadow cloak]

https://github.com/homebridge/ciao/blob/609683b72749eb163ff357e27c4466a6593f907b/src/NetworkManager.ts#L666

src/infra/bonjour.ts uses @homebridge/ciao

but not sure if that codepath is being triggered

[keen spruce]

getLinuxNetworkInterfaces()? ok, that's not something I'd do with neigh ("link" is for that...), but ok. But every 15 seconds? How often do people add and remove network interfaces from their computers usually?

[white galleon]

matches the 15s too!

[keen spruce]

ok, I dropped them a "support request" asking why. Thanks gustavo and bi...etc.

I wouldn't really care if the information "ip neigh" spits out wasn't a list of IPs of other hosts on the attached network. That's very spy-y and with the current efforts of preventing any data extrusion in openclaw... If it was spamming "ip link"...yeah, whatever.

[shadow cloak]

You could just disable bonjour, do you really need it?

[keen spruce]

I didn't even know bonjour was in there. And I don't know what it even does for openclaw...

[shadow cloak]

it's a convenience feature for the client apps to find a gateway on lan

you probably don't need it

[keen spruce]

with how lan connectivity now needs ssl and it's a pain and a half to get a cert for a non-routable IP address...no, I'm not planning on using any of the apps that way. I'm fine with ssh tunnels---they don't need certs.

[keen spruce]

btw, having the weather skill enabled by default and including weather checks in the AGENT.md's hints what to do regularly, we've killed wttr.in...

[inner ledge]

Hey @untold pollen, I've got a few open PRs for Android and wanted to check in before rebasing. The big one is #24239 (iOS invoke parity...). Your device/notifications/camera work superseded my DeviceStatusHandler, and you also introduced InvokeCommandRegistry which changes how commands register, so I'll need to refactor. The 4 remaining command groups (contacts, calendar, photos, motion) are still new though.
Before I rebase, is this something you'd be willing to review and merge if it's clean? I have 6 more PRs that get it closer to iOS parity, including wake word overhaul, talk mode, collapsible blocks in chat, etc. (https://github.com/openclaw/openclaw/pulls/gregmousseau)

[untold pollen]

Hi, thank you for the PR. I'll take a look for sure. Though right now I'd really appreciate some help for the talk mode. I'm happy to take care of the command groups though. I kind of need to manually test everything with a physical device and I have a few things in progress (that I have manually verified) so I'm a bit more inclined to merge those first

[inner ledge]

Great, thanks. I can focus on talk mode. I was actually able to have a useful conversation while driving last night, but could use a couple more tweaks. I also built an APK last week for others to easily test, I'll do that again so it's easy to get others with android devices to test

[untold pollen]

Ah please don't distribute any APKs. We'll try to make it available on Play Store Open Testing soon

Unless users build it themselves sharing non-official APKs is not something I recommend

[inner ledge]

Agree. It definitely needs some of these updates before going to the play store, but that should help speed up development on this

[vocal geyser]

Can I ask whats the vision for the talk mode ? Still only Elevenlabs ?

[untold pollen]

ElevenLabs with a fallback for the device's TTS is what we have so far

Ideally I would love to have OpenAI realtime working

But that's a much larger change and probably not entirely supported by the existing harness

BTW @inner ledge I'm merging a lot of Android stuff lately so there will probably be a lot of merge conflicts. I would really appreciate smaller diffs if possible

[inner ledge]

haha sure. I started with small diffs, and they got progressivly bigger. Let me redo all of them to make your life easier!

[untold pollen]

Start with voice mode for now! I got the rest

[vocal geyser]

yes I know current setup, i was trying to get discord voice chat to be close to talk mode got decent result, but still the android only supports elevenlabs, whats the current stopper for openai stream on android ?

Also is there a reason why it's not unified between macos/ios/android and other plugins?

[sharp cargo]

Just turn on auto generate certificate

And LAN has needed ssl for at least a month. It was like that when I installed it

[dense ingot]

I've got some PR's here:
tui theme command - https://github.com/openclaw/openclaw/pull/26279
migrate plugins.allow when extensions exist - https://github.com/openclaw/openclaw/pull/27766
clarify/persist node exec approvals target in dashboard - https://github.com/openclaw/openclaw/pull/27180

rebased main and those windoes CI checks that were broken yesterday are now passing

[patent bone]

I built a local dashboard to keep tabs on my AI agents — and it’s finally in good enough shape to share.

OpenClaw Dashboard lets you see everything your local AI setup is doing — configs, sessions, cron jobs, API costs, logs — all from one place. It’s completely self‑hosted and never sends data anywhere. No telemetry, no cloud, no weird background syncs. Everything stays on your machine.

It has 10 pages so far:

System Health: Quick read on gateways, channels, disk space, cron jobs, and process health.

API Cost Tracker: Daily spending, model breakdowns, budget limits, and pricing comparisons for 14 models (OpenAI, Anthropic, Google, Meta, Mistral, DeepSeek, etc.).

Session Browser: Browse conversations with token counts and tool calls.

Cron Monitor: See what jobs actually ran and when.

Gateway Logs: Clean, filterable log viewer.

Configuration: Full config with secrets automatically redacted.

Folder Structure: Interactive D3.js tree of your entire deployment.

Plus: a few smaller ones like Memory, Timeline, and Agent Roster.

The fun part: it’s pluggable. Every page is just a folder. Drop a new one into /pages/ and it shows up automatically — no backend edits.

A few more details:

Built with Express + D3.js, no React or bundler.

Dark theme.

Security‑aware: binds to 127.0.0.1, redacts sensitive files, and never reads file contents.

Runs with a single command:

text
npm install && npm start

Repo: http://github.com/joydai2026-del/openclaw-dashboard

If you spin it up, I’d love to hear how it fits into your workflow — or what other pages you’d add.

[hoary pendant]

idk how to bring attention to this new feature proposal, but if anyone has a minute - this thing adds alternative destinations for failures in cronjobs :

• https://github.com/openclaw/openclaw/discussions/29146
• https://github.com/openclaw/openclaw/pull/29145

[empty fjord]

still tryna figure out what pr to make

[modern kestrel]

Is this regarding the iOS app being built or Telegram? I made a PR on Telegram iOS app to integrate Siri support... I understand how the Siri intents work if you need someone to work on Siri in the iOS app.

[untold pollen]

I’m only handling the Android app for now, I think the iOS is also being rebuilt from the ground up by other maintainers

[rich spade]

any idea where I can follow along with this? Just started digging into the iOS app and don't want to start familiarizing myself on an old/deprecated version of the iOS app codebase

[untold pollen]

I can highly recommend just waiting for the apps to at least go out for testing. Once they’re stabilised and all major bugs fixed I don’t imagine it’ll change too much

Make PRs right now would probably just create more work for maintainers

[rich spade]

ah darn, was hoping to start trying to be a contributor/maintainer. But guess I'll have to be patient! Thank you

[limber silo]

small bump on this one, could be an issue if someone accidentally "updates" to 2026.02.15. I know you guys are busy, appreciate any reviews I can get!

[untold pollen]

Reviewing!

[untold pollen]

Merged, thank you ❤️

[halcyon axle]

Hi everyone, I updated my PR https://github.com/openclaw/openclaw/pull/24201 on TPM retry to integrate with pi-ai SDK's retry mechanism. Now detects SDK-internal errors and triggers configurable external retries, mitigating TPM rate limit issues on my end

[viscid lotus]

Hello PR Reviewers/ @untold pollen . I have created a PR to fix the custom provider validation issue coming up when users are using Azure OpenAI endpoints. Based on the api version being used in the code, the request body and the headers needs update. I have segregated the changes for custom providers for azure openai vs non-azure openai endpoints. Kindly review, if this works, Im gonna ask my teammates in office to use openclaw for some of our dev work 😆 :
https://github.com/openclaw/openclaw/pull/29421

[raw crest]

Hi @untold pollen , I reported an issue about skill eligibility checks run on gateway host rather than sandbox container and a PR https://github.com/openclaw/openclaw/pull/29313 was raised for this. Considering it's a high severity issue as it blocks core skill functionality in the primary self-hosted deployment, may I ask if someone could help to review the PR and merge it? Thanks

[sonic ermine]

hey @stable pewter , it would be interesting to consider the possibility of basicaclly making a PR channel for all the prs, and keep clawlaborators as the discussion and info place for coordinating a bit more ambitious code improvements and like design principles ... just a thought , btw i'm sending little prayers every night for you about your huge and massive role in this repo - hope you know that 😉

[viscid lotus]

There is a failing test in the main branch which is failing the PR CIs for all the pull requests:

extensions/feishu/src/client.test.ts > createFeishuWSClient proxy handling > uses proxy env precedence: https_proxy first, then HTTPS_PROXY, then http_proxy/HTTP_PROXY
AssertionError: expected "vi.fn()" to be called with arguments: [ 'http://lower-https:8001' ]
Received:
1st vi.fn() call:
[

• "http://lower-https:8001",

• "http://upper-https:8002",
  ]

Number of calls: 1

❯ extensions/feishu/src/client.test.ts:88:37
86|
87| expect(httpsProxyAgentCtorMock).toHaveBeenCalledTimes(1);
88| expect(httpsProxyAgentCtorMock).toHaveBeenCalledWith("http://lower…
| ^
89| const options = firstWsClientOptions();
90| expect(options.agent).toEqual({ proxyUrl: "http://lower-https:8001…
⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/1]⎯

[sonic ermine]

the best you can do is ignore any un-related tests otherwise you will chase the green light forever , and it's impossible

(my opinion)

[stable pewter]

That’s what we have GitHub for, haha, let’s use that as the source of truth for discussing a given change.

I won’t speak for other maintainers, but if it touches UI/UX, Agent Workflows (DevX), or Docs, then please feel free to @ me in your PR comment.

[viscid lotus]

https://github.com/openclaw/openclaw/pull/29421

Requesting a review for this PR. Azure Open AI endpoints onboarding has been failing with 400 error, this flow only impacts the onboarding of custom provider for Azure Open AI endpoints.

[dry spruce]

Hi — I opened a PR to add a DeJoy channel extension (Matrix-compatible): https://github.com/openclaw/openclaw/pull/29589
Would appreciate a maintainer look when possible; happy to address feedback. Could any maintainers please look at it? Thanks.

[untold pollen]

Merged, thank you ❤️

[dry spruce]

Hi — I opened a PR to add a DeJoy

[untold pollen]

@dry spruce there are 5k+ PRs open. Please don’t tag maintainers.

[dry spruce]

Got it, Thx!

[burnt dagger]

Hi Folks, any feedback on https://github.com/openclaw/openclaw/pull/27275 ? It's a bigger add, but I believe a really big value add with some of the bad press around secrets governance.

[plain perch]

Hey Folks,

I built a LiveKit voice extension for OpenClaw — real-time voice conversations with a dual-agent architecture.

A fast Python agent (LiveKit Agents) handles STT/TTS/LLM latency, while OpenClaw acts as supervisor for reasoning and tools.

Working today: Deepgram STT · ElevenLabs & Hume TTS · Groq LLM · browser voice UI · Cloudflare Quick Tunnel · interactive setup wizard

PR: https://github.com/openclaw/openclaw/pull/29351

Still testing other providers (OpenAI Realtime, Cartesia…) — feedback welcome!

[fallen jolt]

Hi everyone! 👋
I've submitted my first PR to OpenClaw: https://github.com/openclaw/openclaw/pull/29767
This PR adds documentation for running OpenClaw in a macOS Docker container on Linux/Windows, which enables users without Apple hardware to access macOS-only features like iMessage, screen recording, and other node capabilities.
The Greptile review raised some valid security concerns about third-party image sources and VNC configuration. I'm planning to address these by adding appropriate disclaimers and security warnings.
Would appreciate any feedback or review from maintainers. Thanks!

[fallen jolt]

Updated PR #29767 based on Greptile review feedback - added security disclaimers for third-party resources, fixed link paths, and added VNC/credential warnings. Ready for review!
https://github.com/openclaw/openclaw/pull/29767

[shrewd heath]

Hi reviewers/@untold pollen, quick heads up: I opened a bugfix PR 5 days ago and it hasn't been reviewed yet. The issue is still present in the latest release, so I'd really appreciate a look when you have a moment.

Bugfix PR: https://github.com/openclaw/openclaw/pull/24775

I also have two small documentation-related fixes (lower priority):
https://github.com/openclaw/openclaw/pull/25260
https://github.com/openclaw/openclaw/pull/24868

[glacial lodge]

Any maintainers available for review this PR for adding a new provider (StepFun). It's mid-sized but basically following the patterns of other model providers. I can provide an API key to test the model on OpenClaw. Being testing it on OpenClaw for a while. https://github.com/openclaw/openclaw/pull/27211

please DM me for a test API key to make sure it actually works (without going through the sign-in hassle)

[jolly wolf]

#architecture

[drifting fractal]

Hey, I found a bug that destroys the system prompt. Plugins don’t have read access, but they have write access. If they write, the system prompt gets destroyed and this is an easy injection point for malicious code. I’d like to see this get resolved. Can any maintainers help get this through?

https://github.com/openclaw/openclaw/pull/25888

Open to discussion or revisions, just want to see this hole get filled. Thanks.

[outer cloud]

my agent

got an pr accepted he sounds so happy lmao

😂

[final ocean]

stop spamming for a review, Peter has already given feedback just wait

[keen spruce]

to be fair, Peter only addressed the false security implications, not the issue that plugins can replace the prompt but have no access to the old prompt, so they can't edit, but only generate a completely fresh one. Still, @drifting fractal, please remove that security claim from the PR. It has already been disproven and just muddies the waters. I wouldn't even have looked further into the PR the way it's now. Yes, you found a logic gap (assuming my 2-minute look at the calling code didn't miss something), and if you clearly state it, the PR might get noticed again.

[drifting fractal]

Alright, what should I do differently?

[fossil adder]

we get flagged for CVEs that don't apply in our products, it does take some thinking. Might be nice to have a CVE dashboard if that isn't considered a weakness. E.g. found date, severity, state (pending review, reviewed with estimate, resolved in tag). Sorry I'm here for something else and couldn't resist 2 more pennies to the mix

🔍 Workflow Pattern Exploration — Human-Supervised Orchestration

I'm moving towards a structured workflow pattern on top of OpenClaw where I vibe code with a GPT-ish agent and it does MCP to my OpenClaw server. Would love feedback on whether this aligns with project direction, or interests, or somehow might seem rude.

Concept

1. Low-cost conversational planning layer

   • Discuss changes
   • Pull GitHub context as needed
   • Refine improvement ideas
   • Minimize API token burn

2. Explicit human approval

   • No execution until approved

3. Structured backlog submission

   • Approved items enter a tracked execution queue
   • Status queryable

4. Execution agent runs against staging

5. Human review + explicit promotion to production

Goals

• Reduce unnecessary token spend
• Add guardrail clarity
• Introduce lifecycle structure (plan → stage → promote)
• Make OpenClaw viable for production dev workflows

Open Question

Would something like this:

• Fit as a "Supervisor Mode" concept?
• Be better implemented as an MCP-layered workflow?
• Or live outside OpenClaw and call it via API?

Curious how maintainers think about structured lifecycle orchestration on top of the current agent loop.

[keen spruce]

not just CVEs, but any kind of security fix. What's almost always missing in those PRs/reports is (a) What's the attack vector? and (b) How would a fix discriminate between legitimate user action and an attack? (c) What's the value balance between overblocking legitimate user actions and underblocking attacks? (d) Is this a pebble in a wide-open barn door? i.e. How many other ways of doing bad things does an attacker that can execute this attack have?

For example, "You can type rm -rf * into bash" is a security finding, no arguing here. However, (a) "attacker already has control over the system", (b) "no way", (c) "99.9999+% of blocks would be overblocking", and (d) "yes, very many"

[jolly wolf]

fwiw a lot of the security fixes and reports aren't public and they do have that, yall see the ones that don't follow our existing security policy

[keen spruce]

Can someone please re-open this? https://github.com/openclaw/openclaw/pull/11444 It's not a plugin, it's the infrastructure so people can add seacrh providers as plugins.
(And if it gets rejected for other reasons, fine. I'd just hate it getting killed by a bot misreading the intention.)

[jolly wolf]

1. yep reopened!
2. check who added the label that triggered the close not the bot sending the close snippet

[wooden prawn]

Is there any chance somebody can look at this for me? I would like to move away from using the filesystem for storing data and move to using the postgres database that I think most of us have alongside claude. You can set OPENCLAW_DATASTORE=fs or OPENCLAW_DATASTORE=postgres and then set the OPENCLAW_STATE_DB_URL to your postgres url and it'll intercept all filesystem operations and move them into the database instead. This would make it more resilient from a reinstallation point of view as well as making it easier for people running in containers where you don't need to then setup extra persistent volumes docker or pvcs in kubernetes.

It fails some test, but I'm unsure the reason why, it appears like it's a timeout, I've rebased it a couple of times, but if nobody likes the idea, It'd be good to hear some comments if there are any?

https://github.com/openclaw/openclaw/pull/29008

[keen spruce]

I saw it after I wrote the message...and then decided not to edit it. 😜 I mean, you sounded pretty human, but some people still have their doubts...

BTW, if the code in the PR is good, it might be worth merging. The author has kept it alive for ages. I can dig into the code if you want; I think I only had a cursory glance weeks ago.

[wooden prawn]

the tests I think are a little flakey, sometimes I get failures in code that I never touched. Perhaps something was merged that isn't quite right

[slow pagoda]

I made a scanner that follows Peter's security visions. I like it works great.

[keen spruce]

Oh, my! That's a big one. Looks good from my end (note: green name, not golden one), assuming your design decisions are valid, you found every reference to file access, and it actually works.

[wooden prawn]

my next PR will be about eliminating the gateway token and replacing it with a jwt user auth 😉

and after that, onboarding wizard in the app, no terminal required

They're not that scary or big though, the postgres datastore branch was annoying because it changed some functions use of async/await and it's just lots of files where it's either added or not. But no real code changes, you know?

[jolly wolf]

For what it’s worth, this will likely either never merge or be a long drawn out process because there are so many third-party things as well as first party things that would all have to be updated

[wooden prawn]

from what I can see, the jwt is a drop in replacement for the gateway token, it's just a string one way or the other. Which means you just update whatever gateway token you've supplied with your jwt token instead. A further update would be to let the admin user create a jwt token that never expires, so nobody has to worry about their linked services suddenly dropping. Can you see any reason why it would be a problem?

I think the benefits are quite obvious though, it would make the system more secure, allow for users to be created on a single openclaw, and allow for permission segregation based on what token you have. So whilst making the foundation more secure, it gives further future options

[keen spruce]

I'm pretty sure the current pairing using device keys system already does all this. All that is missing is a pre-mapping from tokens to permissions, so you can hand out different tokens that already come with permissions and don't snap to whatever is approved on first contact and bound to the device key. Users is a whole different joblot. At the moment, there is only one user---not because there couldn't be more, but because access to the Agent gives you deep system access already. Only if/when agents are safely locked away in containers would users even make sense. But, to be honest, I think it makes more sense to separate users by having different gateways on different operating system users than locking in agents and trying to prevent them from using each other's resources (think Discord bot tokens, Telegram accounts, and so on).

btw, there is a PR that adds single-use gateway tokens for pairing devices. I find that one interesting, as it removes a big attack surface. Escaped tokens would no longer be useful.

[glacial lodge]

Not to spam but just trying to get some attention here. The Step 3.5 Flash model is the second most used model (as of Feb. 28) and 4th most used monthly for OpenClaw on OpenRouter. https://openrouter.ai/apps?url=https%3A%2F%2Fopenclaw.ai%2F

This PR simplies the onboarding experience for new OpenClaw users who wish to use this model to get started.

The PR is mid-sized but it follows the patterns of other model providers in onboard auth handling and docs with no deviations. All tests passed and greptile rates 5/5. If needed, I can provide an API key for testing the actual model integration.

Appreciate any comments or pointers from maintainers.

https://github.com/openclaw/openclaw/pull/27211

[wicked patrol]

Hi Team,
I found a bug that affects the system prompt integrity. Plugins don’t have read access, but they do have write access. When they write, the system prompt can get overridden, which creates an easy injection point for malicious code.

I’d like to see this resolved. Could any maintainers please help review and move this forward?

Issue: https://github.com/openclaw/openclaw/issues/30402
PR: https://github.com/openclaw/openclaw/pull/30407

[raw crest]

Hi everyone, I’ve noticed there are currently over 5k open PRs in the repository. It seems the community might be facing a significant backlog. Are there any existing plans or community-driven solutions to manage this volume? I'm concerned that a long-term backlog might discourage new contributors and create a bottleneck for the project's growth.

[vocal geyser]

this was discussed yesterday in this channel #clawtributors message

[wooden prawn]

yeah I think I'm more interested in replacing the giant global admin key with something that's controlled, revokable, and has a lifetime that could expire than creating users per se. It's just that if we replace it with JWT, then we get user login for free, which means the system basically becomes secure by default instead of "if you have this string, you're admin". Because if the gateway and the control panel required you to create a user and login. You wouldn't have exposed control panels without login control all over the internet. So from that perspective, even though it's "marginal" in terms of utility. It's actually pretty huge in terms of security, you know?

Having a JWT token that expires every 2 hours for example would mean your panel and gateway can't be used until you login again, ot refresh your token, or similar. Even if there is just one user, I think this is just a basic thing we need if we want to make a secure platform.

[hoary pendant]

Have been using it in my bot for a couple of days. It is an absolute improvement - reduced noise in normal channels and centralized all faults into a single one. Life saver!

[dusky hollow]

Hi everyone — I’m @inkdust2021 on GitHub. I built VibeGuard, a local redaction layer that replaces configured secrets/PII with reversible placeholders before prompts/tool history are sent to LLM providers, then restores them locally (and restores placeholders in tool args before execution).

• Project: https://github.com/inkdust2021/VibeGuard
• OpenCode integration (ecosystem docs entry merged): https://github.com/anomalyco/opencode/pull/15464
  I’d like to contribute a similar capability to OpenClaw. Does OpenClaw have any hook/plugin point to implementoutbound redaction without core changes? If not, what’s the preferred path here (issue first, design discussion, or asmall draft PR)? Happy to follow your contribution guidelines and adjust config/placeholder format/tests/docs as needed. Thanks!

[shadow cloak]

Is it intended default behaviour in the latest release that chats from the main session from the webui doesn't resolve as an "owner" chat so tools like cron and gateway are disabled?

feels like a breaking change and commands.ownerAllowFrom isn't updated yet in the latest mintlify docs, I have to add

  "commands": {
    "ownerAllowFrom": [
      "openclaw-control-ui"
    ]
  },

to the config

[shrewd heath]

I just noticed the merged fix for the later duplicate issue (commit c58d2aa by @wide geyser ) does not fully cover the earlier fix I made in #24775.

I opened #24775 earlier for #20320 and it still adds broader edge-case handling plus additional tests beyond what landed in c58d2aa.

PR: https://github.com/openclaw/openclaw/pull/24775

I’ve rebased it on main and resolved conflicts. Could a maintainer please have a look?

[tall prairie]

https://tenor.com/view/pull-request-pr-jira-bitbucket-merged-gif-27373367

[shrewd heath]

Thanks a ton @tall prairie , really appreciate the quick reaction! 😊 Can I please also have the clawtributors role now? 😇

[tall prairie]

I think mods give them out

[shrewd heath]

Oh sorry, just saw I misread the server FAQs, are the community staff the mods? Just noticed they probably have to be the ones, as all other roles don't match. Maybe it could be useful to also change the FAQs to say "community staff" instead of mod 😊
As soon as one is online I'll ping them

[viscid lotus]

https://github.com/openclaw/openclaw/pull/30748

Hi PR Reviewers, This is a small PR to support bangla(Bengali, India) locale in the web UI. Would be great if someone could do a review. Provided reference PR as well, took reference of the German locale addition PR.

Also, widingmarcus-cyber has created a PR to unblock a failing test blocking the CI for every PR: https://github.com/openclaw/openclaw/pull/30750
This would unblock a lot of clawtributors whose CI is failing due to the same test. 😁

[dusky hollow]

can i submit pr?

[final ocean]

Main is passing tests fine

Closing PR, we will resolve any bad tests

[lean nova]

Hiya! can anyone explain not planned to me?

[slow pagoda]

They need stricter linting focused scope and comprehensive test results. Keep diffs small & single-purpose
Add tests for anything you touch (especially new/fixed behavior)
Explicitly reference prior reviews/feedback in the description
Run/fix lint early (your import reorders show you did)
Use clear, incremental commits
You can find almost everything in doc's. This is what I followed to get a clean 5/5 Greptile confident score.

[final ocean]

Please stop submitting PRs for a failing tests. We will resolve any and rebase your PRs

[viscid lotus]

I had raised a different PR(https://github.com/openclaw/openclaw/pull/30748 Web UI locale support addition) where that unrelated test was failing consistently. And it is completely unrelated.
Noticed that it was failing for most PRs raised recently so informed here as one of the contributors raised the PR to fix the failing test. Will keep in mind going forward. 👍

[shrewd heath]

Did you submit a (large) feature? If so, they mentioned they are focusing on stabilisation for now, so that could explain it. Otherwise I think you should include a link to the PR/issue probably makes it easier

[dusky hollow]

well,I want to contribute a new feature.But I see the CONTRIBUTING.md said I should ask here or the github discussion fisrt,but how can I know if I'm allowed to submit PR?

[sharp cargo]

wtf why did you write exactly the same post as someone else for exactly the same issue but open a new pr for it

Edit: even weirder, this issue/PR is completely unrelated?

[shrewd heath]

Write here or in a github discussion about what kind of feature you want and why, then you'll be able to discuss all the details with the maintainers (like whether they want the feature or not, and how it would be implemented). If they accept the feature, you'll create an issue with the details you discussed, at which point you can submit a PR to implement that feature. Before that, submitting a PR to implement a new feature will probably not get accepted. Just keep in mind they also say on the PR page that OpenClaw is in the stabilization phase, so it may take a while for a new feature to get added.

[slow pagoda]

I want to give my Windows PC Claws 🦞
Noctis said there's not a PR for this yet. I'm just waiting for the stabilization period is good then put it forth unless the team implements that through this stabilization period then that'd be dope AF! Thanks for the hard work!

[shrewd heath]

There is windows support already if you use WSL2

[slow pagoda]

I don't unfortunately.
So I guess I'm going to dual boot. I miss Linux.

[shrewd heath]

Why not simply use WSL2 then? It's much easier than setting up dual boot 😁

[keen spruce]

Is the "mudrii" who's running the "Triage (community PR sweep)" bot over GitHub here? My notifications are stuffed full of those comments, and so far, not a single one of them has had any useful information. We can all see if a PR is mergeable or conflicting, and if the CI failed or not. GitHub already shows that natively. And listing other PRs that are already linked by earlier comments or in the PR's text itself isn't useful, it only creates more work when reviewing, as you now have to check if those are additional links or not. Dial that thing down, and make it edit its post, not add new ones, at the very least, please.

[slow pagoda]

Nice, I'll do that then! Hellyeah 🤘💪💯🦞
Thank you 😊

[shrewd heath]

Give it a try, it's really easy. You're welcome! 😊

[hoary pendant]

🙁 my PR got closed in favor of a PR that doesn't implement my PRs features ... in mine https://github.com/openclaw/openclaw/pull/29145 I could set a global default for all cron failures to go to a specific channel, but in https://github.com/openclaw/openclaw/pull/24789 that superseded it, I have to specify each cron's failure individually ... 🙁

mine was closed my a maintainer ... not gonna help checking it, it was killed.

no, the PR that got merged instead was much wider, included web UI changes, and generally looks like a lot more slopp-ier

ehhh.... guess I'll just need to spend another couple trillion tokens adding my functionality to the meh PR that got merged

well, just based on the PR number, it was done before mine.

as-if the AI will be able to find similar ones 🙂 especially since one used the terminology failureAlert, and the other used failureDelivery 🙂

not enough context to know everything about all the 5k PRs out there

yeah, I have the github mcp, similar to gh cli

I mean when you want to start work on something, and you have 5000+ PRs open. You don't have enough context in the LLM to find if there is anything that already implements what you want to start working on.

My method of picking what to work on usually starts with me being annoyed at some existing way that things work, and I just start fixing it.

like it was annoying to receive cron failures into the channels where cron results/successes were supposed to go ... so I worked on that.

or it was annoying that telegram DMs have topics just like groups do, but there is no support for it. so I worked on that.

or it was annoying that the AI decided to put delivery names for telegram like "channel/topic" instead of the actually working "channel:topic", so I worked on that.

for me too! I reverted that whole commit that introduced this stupid thing.

I also have to rip out the signal pnpm dependency that uses "git" to get its code, and to rip out the a2ui build thing. both of those just never work, so I just patch them out completely from my claw.

I actually spent some time changing the way my local pnpm works, added a whole bunch of flags that it may or may not need, to try and not have to do these patches. but this monolithic monster is extremely fragile, and I have been upgrading it about four times already, each time it takes half a day to fix the newly introduced "wtf" things.

I upgrade to git main usually, plus my own patches, which I manage via quilt. Most of which are my pending PRs.

I have a custom way of installing it, using a NixOS module.

I have a systemd unit that does git commit every half an hour for all files, and uploads that to a remote location. But having the ability to "build from scratch" using the exact same version and settings that "worked before" is a godsend. Especially for something that breaks every other day.

even when it works, it barely works 🙂

it loves bombing its own openclaw.json with settings that don't exist from time to time

and the "doctor" solves things by just creating an empty openclaw.json half the time

don't think there is that much difference between the dev and non-dev versions tbh

here I go ... spending more time to implement what I already had implemented and working 🤦‍♂️ ... https://github.com/openclaw/openclaw/pull/31059

I kinda like the whole "greptile + codex leaving review comments" game, but it does take quite a long time to get to a finish line while hopping around them each time they pop with a new insight

Good night! its not just the CI tests, codex and greptile keep finding new nitpicks ...

[marsh seal]

Hey 👋 I have an open PR that could use a review when someone has a moment: #29985 — adds api.resetSession(key,
reason?) to the Plugin SDK.

It extracts the existing sessions.reset logic into a shared helper so plugins can programmatically reset sessions with
a 5-second cooldown to prevent loops. 18 new tests, no regressions, all bot reviews addressed.

This also unblocks #30452 (flush-then-reset compaction mode) and complements #30764 (session lifecycle hooks) — a few
people have been asking for programmatic session management.

Happy to address any feedback. Thanks!

[oak shard]

Hello, I threw up PR #23910 from last week that went stale today. Would love to re-ping it back in the pool and get a second pair of eyes on it.

I think it’s a high impact quick win with low overhead. Implements a session hook (detecting changes in models, labels, verbosity, etc.) without having to make a whole cron job.

I've been using this in my own projects extensively for organization, db sync, and cost tracking. Greptile gives it 5/5.

PR: https://github.com/openclaw/openclaw/pull/23910
Discussion: https://github.com/openclaw/openclaw/discussions/24877

[viscid lotus]

Hello PR Reviewers, Raised this PR https://github.com/openclaw/openclaw/pull/30748
This is for adding Bangla locale to the web UI. Took German locale addition PR as reference. Would be great if someone could review this.

[icy plover]

i am working on a fix for tui streaming issues, and i've noticed that regularly my branch is failing CI because main is broken, i'm wondering how this is happening that PRs presmably passing CI, then merging, but cause other PRs to fail CI for unrelated code. It has happened a couple of times that my branch review progress gets stalled by this, CI checks failing in areas of code that I didn't touch, and I check the PRs page and see all recent PRs failing.

Presumably these are green PRs when they merge, how are the breaking main?

[final ocean]

we regularly make many changes at once so ci can fail on main

[icy plover]

Also for the issue I'm working on, it seems that the TUI frontend has a couple of issues

• Docs say streaming reasoning is an option, UI select only shows on or off. If you type an invalid value, it shows stream is an option. Nit, quck fix
• Streaming reasoning doesn't seem to work at all when toggled. The agent is aware of the reasoning text, but says it is hidden in both "on" and "stream" modes
• Streaming responses that involve intermediate tool calls causes the text of the intermediate tool calls to be erased, and you can't see the full stream of output.

I really like the TUI so i'm hoping I can get these fixed. I threw claude 4.6 at it and i'm thinking of trying a clean-room implementation with codex to compare them. Gotta admit I don't really understand the internals much though, just the broad strokes.

[final ocean]

its not a hard rule for us internally to make it fast

pass*

[neon gorge]

first PR here https://github.com/openclaw/openclaw/pull/29590

bridges after_tool_call to internal hooks so ~/.openclaw/hooks/ handlers can actually see tool executions. 50 lines, 3 files, follows existing patterns

ref discussion #20575. would appreciate any eyes on it

[final ocean]

self assigning as im looking into all the hooks tomorrow

[neon gorge]

thanks

[dusky hollow]

Thanks for your reply!In fact, I writed my idea here but I haven't received any feedback.Should I say it again?

[dense forge]

I developed a channel connect my instant messenger to openclaw, how to integrated into the source code of openclaw like twitch so user can install using a command like this: openclaw plugins install @openclaw/twitch anyone can help?

[shrewd heath]

Maybe try github discussions? I think it's harder to get lost that way

[dusky hollow]

I haven't received feedback in github discussions too 🙁 Maybe I should post again

[shrewd heath]

Just give it some time, as I mentioned features are not a priority right now. And if you want, you can still extend OpenClaw even without integrating it into the codebase itself if you want to have your feature earlier, see: https://github.com/openclaw/openclaw/issues/5799

[dusky hollow]

Thanks so much for your reply. I’d like to create an extension for OpenClaw. Does the OpenClaw repository have documentation on how to display or register extensions? My English is not good, thanks for your patience.

[shrewd heath]

To be honest, I don't know, maybe someone else here knows? Otherwise you can always try to ask your OpenClaw if it can tell you or do it for you 😁

[dusky hollow]

LOL, thanks so much

[tawny vine]

As per FAQ, I am pinging mods to get the clawtributor role! Here is one of my PRs: https://github.com/openclaw/openclaw/pull/30853 - @dusky blaze @edgy plaza @winged socket @patent berry

[dusky blaze]

Hey Mawen, the PR has not been merged yet. Once it is, ping us again and we'll be happy to add the role 🙂

[hallow badger]

Requesting the Clawtributors role @dusky blaze

https://github.com/openclaw/openclaw/pull/20318
https://github.com/openclaw/openclaw/pull/18201
https://github.com/openclaw/openclaw/pull/18188
https://github.com/openclaw/openclaw/pull/17136

[tawny vine]

https://github.com/openclaw/openclaw/pull/27936

[dusky blaze]

@tawny vine - thank you 🙂 Done!

[dusky blaze]

Done 🙂

[hallow badger]

Thank you!

[keen spruce]

Could someone with GitHub super-powers please get rid of this garbage pile? https://github.com/openclaw/openclaw/pull/11970 Someone PRed their working branch and just won't notice...
Thanks in advance.

[untold pollen]

Closed 🫡

[sour condor]

hi guys!
this is coral and I planed an online event for openclaw devs projects sharing, does someone know how could I put the event to the event dashboard in this discord?

[viscid lotus]

https://github.com/openclaw/openclaw/pull/30748

Hello PR Reviewers, Could someone please review this locale addition PR? I took reference of the German locale addition PR and have also updated the i18n unit tests 😁 only the text conversion is AI assisted.

[hoary pendant]

Is there some mechanism for PR closure appeals? my PR got closed in favor of another PR that implements less features, but does implement features, and the PR that I made to add back the missing features got closed for the reason of "we are not adding any new features right now per the vision" ... eh? what? Why? https://github.com/openclaw/openclaw/pull/31059

[shut forge]

This commit overwrote "2026.3.1" in CHANGELOG.md, but it doesn't look like openclaw@2026.3.1 has been withdrawn. 🤷‍♂️
https://github.com/openclaw/openclaw/commit/6ba7238ac6239e79e9849af4cc8c2161330b1d79

[drifting fractal]

@keen spruce, updated the PR according to your suggestion. Also changed the title per your request. Lmk if you need anything else. https://github.com/openclaw/openclaw/pull/25888

[inland field]

https://github.com/openclaw/openclaw/pull/30746#issuecomment-3982864361

may i have the role sire

[thorny coyote]

Hello maintainers This is a simple and straightforward fix to timestamp of logger as docs advertised. This is quite annoying bug that both confusing humans and agents while debugging. Thanks for your time
https://github.com/openclaw/openclaw/pull/28434

[final ocean]

about to merge

merged

[thorny coyote]

Thanks dude

[zenith stone]

nice

[formal berry]

If I'm updating a small item in the docs, but now the linter is throwing a bunch of issues with other parts of the docs, would you rather I work through all the issues or just ignore the downstream problems so the PR maintains its small scope?

[zenith stone]

i think it's better to ignore the issues

or make separate PRs for the other parts of the docs

[formal berry]

Now I have to remember how to work with git locally and squash lol. I'm so out of practice

[zenith stone]

you got this!

[shadow cloak]

Apologies for the bump, but I thought I'd try one last time before I just close the PR: https://github.com/openclaw/openclaw/pull/28599

Basically, any interest in using bwrap as a tool sandboxing backend instead of docker?

[zenith stone]

wow, that's a large PR

might be good to ask whether this is something that the maintainers would want

[shadow cloak]

It was a bit of work to get bwrap to work as a sandboxing backend instead of Docker yep, though a large chunk of it is AI generated tests

[shadow cloak]

where do I ask?

[zenith stone]

here!

[shadow cloak]

okay perfect 😂

[zenith stone]

or #architecture maybe

[shadow cloak]

I'd also be happy to maintain this specific feature, I've deployed my fork for my own setup atm

[tall prairie]

u should be good now

[final ocean]

Not at this stage

[shadow cloak]

Okiedokes

I'll close for now to not contribute to the sea of 5000 open PRs

[zenith stone]

i just wanted to bring up a config fix I made a couple days ago: https://github.com/openclaw/openclaw/pull/27867

currently, when plugin validation fails with INVALID_CONFIG, loadConfig returns {}, dropping all user config, including unrelated core/channel settings.

my PR runs core validation if plugin validation fails with INVALID_CONFIG, preventing unrelated core settings from being lost if core validation passes.

[viscid lotus]

Hi All, Am i eligible for a Clawtributor role for this?
https://github.com/openclaw/openclaw/pull/29421
If yes, shall I have it? 😁

[timber beacon]

Requesting Clawtributor role @raven kernel
https://github.com/openclaw/openclaw/pull/18496
https://github.com/openclaw/openclaw/pull/18555
https://github.com/openclaw/openclaw/pull/29091

[viscid lotus]

Thank You! 😇 🙏

[lethal lodge]

Requesting Clawtributor role @raven kernel
https://github.com/openclaw/openclaw/pull/30978

[raven kernel]

welcome! and congrats on landing code in main!

[lethal lodge]

Thank you 😁

[zenith stone]

Yay

[formal berry]

This is a very small change in the docs that I think is extremely helpful for newcomers like myself. Essentially, it turns the recommended update path into a single command that can be copied exactly like the source install below it. When you don't have a lot of terminal or programming experience, you're not sure how to parse a flag like --no-onboard simply:

https://github.com/openclaw/openclaw/pull/31858/changes

[jolly wolf]

I’d rather not make skipping onboarding even easier for people, especially if they don’t have experience……

[formal berry]

Even though that's buried in an update step?

[jolly wolf]

I could kind of see a point if this was targeting more experienced people, but those experienced people could just add the flag

[jolly wolf]

Yeah, because often times new features will be included in the onboarding stuff, and our onboarding flow shows you what your current settings are, it doesn’t overwrite them

[formal berry]

That does make sense to me. I was seeing it from the perspective of making updating using the shell script route easier

[foggy dragon]

Any way I can contribute (open requests, bugs, etc)? Im a seasoned dev with 25 years experience from front end to data stuff.

[nocturne plume]

I have a PR open that fixes an auto restart loop when using IRC channels an issue that multiple people have reported in bugs. I have tested it on my system another PR commented seeing good results with it. Its been open for a little while (5 days) and is isolated to IRC integration.
If a maintainer has a chance to look/review and possibly merge it would be much appreciated. I have updated the PR so maintainers can update the PR commits.

https://github.com/openclaw/openclaw/pull/26918

[nocturne plume]

There is a second IRC issue that was discovered in the fixing of this one and will hopefully have a PR up for that too today. Its related to this issue but a different issue.

[sharp cargo]

Just look at the GitHub repo lol

[jolly wolf]

https://x.com/steipete/status/2028541411667148852

[glossy yew]

@steipete via Twitter

Peter Steinberger 🦞 (@steipete)

Another milestone. 🫶 to all Clawtributors!

**💬 6 🔁 1 ❤️ 27 👁️ 536 **

FxTwitter · Mar 2, 2026, 6:42 PM

[foggy dragon]

Oh man 4k open issues 😱

Im gonna need a bot to figure out which one to help with.

[vocal geyser]

It's a trap 😄

[marsh seal]

Hey, I know things are overloaded but is there anyway someone review and merge https://github.com/openclaw/openclaw/pull/29985 ? It adds api.resetSession() to the plugin SDK with a shared resetSessionByKey() helper and 5s cooldown guard — plus fixes a pre-existing bun-only flaky cron test.

[vocal geyser]

legit question to the more experienced Clawtributors, when you stop trying to make the vibecoded stuff perfect ? And just say this is good enough

[pearl lintel]

Has anyone succeeded in get the browser relay extention to work ?

[sharp cargo]

It should be good enough that I wouldn’t ever tell someone I didn’t write it as some sort of excuse.

[foggy dragon]

When it does what you want, and you have passing unit/integration tests.

[vocal geyser]

well i got it working but tried to satisfy every single codex-review concern and i kept going and going i solved all concerns but then during manual testing i found something fixed it and then it started with more new concerns and scope kept increasing, And after 4 days of working on this PR I don't even think it will even get a maintainer review/consideration.

[mental rivet]

I opened a PR so that you can natively authenticate with Codex for openclaw models auth command, so that the command suggested by the docs (openclaw models auth login --provider openai-codex) works.

https://github.com/openclaw/openclaw/pull/32065

In the future, will add the other providers in the onboard command onto the login sub-command.

[empty jolt]

hey @raven kernel , requesting Clawtributor role 🙏

merged PRs:

• #31090 fix: handle CLI session expired errors gracefully
• #31088 fix(gateway): support wildcard in controlUi.allowedOrigins
• #27276 fix(daemon): stabilize LaunchAgent restart and proxy env passthrough
• #24134 fix(config): keep write inputs immutable when using unsetPaths
• #23379 fix(stability): patch high-severity runtime regressions
• #23284 fix(telegram): prevent update offset skipping queued updates
• #3628 Fix text attachment MIME misclassification

https://github.com/openclaw/openclaw/pulls?q=is%3Apr+author%3Afrankekn+is%3Amerged

[marsh seal]

Any clawtributors with experience able to give me some advice? How frequently should I have my bot check and maintain this while I wait for human review? Given that it looks like it would take a long time I built a fork to push from so that I can keep working around the feature, but now I need approval for that https://github.com/openclaw/openclaw/actions/runs/22593710569

[twin mirage]

ha all 🙂

[modern kestrel]

I guess I'm supposed to request the CLAW badge too?

https://github.com/openclaw/openclaw/pulls?q=is%3Apr+author%3Ahuntharo+is%3Aclosed

Merged PRs of note:

#26933 - L - Fix telegram webhook hang on restart
#25732 - S - Improve telegram webhook logging and docs in config UI
#22019 - XS - Reduce app-specific docker size by ~50%
#21316 - XS - Improve Sonos skill guidance when discovery fails
#17774 - L - Timezone handling fix on Usage report

Wait... did I just give it to myself? That's all it took? Or does that mean someone gave it to me before?

[tall prairie]

You belong to me now

you've been knighted!

[twin mirage]

requesting Clawtributor role 🍻

Merged Prs:
• #21074 — security(web_fetch): strip hidden content (prompt injection fix)
• #21859 — fix(logs): TZ env var + Windows timestamp fix ← vandaag gemerged!
• #23065 — fix(slash): persist channel metadata from slash command sessions

[marsh seal]

Thanks to whoever approved my fork. All CIs passed. Any clawtributors with experience able to give me some advice on how frequently should I have my bot check and maintain the PR while I wait for human review?

[vapid pawn]

Heads up Lobster Crew 🦞**

main currently has a few regressions and flakes that are blocking CI across the repo. I’ve submitted four strictly scoped PRs to resolve these and unblock the lobster tank:

1. #32122: Fixes a blocking duplicate test helper in Synology Chat.
2. #32159: Fixes an intermittent skill-scanner cache invalidation flake.
3. #32165: Fixes a non-deterministic mock state flake in Cron tests.
4. #32119: The primary fix for the pnpm/bun hardlink global install bug.

I’ve verified the combined stack as 100% Green locally. Merging the first three will clear the CI "red" status for all active PRs.

[hazy jasper]

@celest merlin @joshavant @sebslight I've prepared a clean-up stack to unblock current main regressions/flakes (#32122, #32159, #32165) alongside the pnpm hardlink fix (#32119). Verified everything 100% green locally.

[viscid jay]

@vapid pawn, please don't ping Peter for issues, use #1459642797895319552 or #users-helping-users if you need help, use #clawtributors to discuss PRs, or use any of the many other channels in this server as they're intended.

If you have a problem with the Discord specifically, use #report or DM @jolly wolf
-# Your message was reposted above without the ping active for the sake of conversation.

[jolly wolf]

please dont open prs to fix ci, treat it as a suggestion

[raven kernel]

holy crap.

[mental rivet]

Sorry, just bumping this, now that I've gotten CI to work smh:

I opened a PR so that you can natively authenticate with Codex for openclaw models auth command, so that the command suggested by the docs (openclaw models auth login --provider openai-codex) works.

https://github.com/openclaw/openclaw/pull/32065

In the future, will add the other providers in the onboard command onto the login sub-command.

[full hearth]

was there an update to the browser relay

[left mango]

Can someone please close this issue? It has been fixed already. Thanks.
https://github.com/openclaw/openclaw/issues/32088

[viscid lotus]

A huge shoutout to the folks who fixed all the flaky tests/test regressions in the main branch. I am finally able to witness a green on my PR's CI 🙏 😇

[mental rivet]

Agreed.

Had to piush like 10 unnecessary commits because the tests were kinda broken

[oak fiber]

make a pr and ci failure with somthing not relate to my code, how to retrigger ci without push again？ does any have idea with it？

[mental rivet]

@jolly wolf it’s NOT a pr for test fails 🙏

[jolly wolf]

looks like a pr for test fails gonna ban

[mental rivet]

It isn’t
It had a commit for one but it was done by someone else and merged in 😂

[unborn isle]

Not sure what is the correct way to contribute and to ping about PRs, I created a PR to make blockStreamingBreak more flexible so I can make it stream block in DMs but in channels with multi-agent non streaming: https://github.com/openclaw/openclaw/pull/31289 . Let me know if this is the correct channel or I should go to other places 😄

[heavy rune]

I saw the recently added Diffs extension which looks like a great idea. In practice, I'm finding it's user message scoped injection of the When you need to show edits as a real diff, prefer the diffs tool... DIFFS_AGENT_GUIDANCE is resulting in a lot of "Noted on diffs - I’ll use it as the default for real edit diffs going forward." messages across chats. I wanted to check on the thinking around these instructions being injected before_prompt_build rather than this being implemented as a skill or similar? If it needs to be here, it's possible this wording needs some tweaking to avoid it being interpreted as a new change that the agent needs to comment on? Cc/ @white galleon

[white galleon]

happy to iterate and improve; have you tried removing the prompt and running to make sure it;s what's triggering the messages you are seeing? which model are you using?

[heavy rune]

seeing similar sort of responses with sonnet 4.6 and codex 5.3. Mostly I just searched for "diff" and noticed that agents guidance message getting loaded in a lot in my session files.

Which struck me as odd since none of these chats were code / change related that would trigger a diff.

[white galleon]

will fix that

[keen spruce]

um, the user prompt? That would be bad, tbh. Throwing that long list of text into each message the user types would be ... um ... undesirable. But it looks like it should be going on the system prompt. Not needed, in my opinion, if the tool description is good enough
But, tbh, all those prompt injection callbacks are too similar for me to keep apart. niming isn't ideal in that area.

As it has gone "stale", and I don't really want to mess with the branch when it still is mergable just to get barnacle off my back...: https://github.com/openclaw/openclaw/pull/27028 a one-liner QMD-upsteam-bug workaround. Has been running here for a week, so it works.
Yeah, sorry for spamming "notice me"...
PS: And one for updating outdated documentation: https://github.com/openclaw/openclaw/pull/26860

[heavy rune]

Thanks. I've been testing and playing with this a bit today and my suggestion would be to migrate the agent guidance to a diffs skill within the tool/plugin. Proposed PR if it helps: https://github.com/openclaw/openclaw/pull/32630

[hearty condor]

Bug: /acp slash command ignores inline arg, shows menu anyway

When typing /acp close (or any action inline), Discord sends the interaction with action = null because the field uses static choices — so OpenClaw always falls back to the button picker menu.

Root cause: buildDiscordCommandOptions registers the action arg with static choices (17 options, under Discord's 25 limit), which forces Discord's dropdown UI. Users can't bypass it by typing inline.

Fix: Register action with autocomplete: true instead of static choices. This lets users type freely, Discord passes the value through, and resolveCommandArgMenu correctly skips the menu when a value is present.

Cant just type /acp close in discord, it still shows the selection menu instead of bypassing it straight to closing the acp sub agent

[keen spruce]

Please report issues on GitHub (https://github.com/openclaw/openclaw/issues or https://github.com/openclaw/openclaw/security). The only exception would be recent breakages/regressions that can be tracked down to a specific commit/merged PR.

[sharp cargo]

hey sweet someone (peter?) fixed the tool result compaction code

[zealous plover]

I have a simple Docs PR if someone wants to take a look for me.

https://github.com/openclaw/openclaw/pull/32695

[proven moon]

Hi, I have this small pr as a warmup. Its related to heartbeat skipping triggers on request in flight errors. Added exp backoff for hb rates bigger than 1h

https://github.com/openclaw/openclaw/pull/31638

[drowsy light]

Hi all, this PR should still be tested and ready to go, it fixes a bug preventing onmessage from working in Mattermost channels: https://github.com/openclaw/openclaw/pull/20051

[silver crown]

Hi everyone. I needed a good way to capture messages as they persist. currently the memory is parsed from jsonl file post session. but we can capture the message as they happen using a after_message_write hook
here's my PR - https://github.com/openclaw/openclaw/pull/32855

[oak fiber]

Hi, sorry for bothering, i have a minimal and surgical PR ready for review, which unify console timestamps output. hope someone could take a review , thx: https://github.com/openclaw/openclaw/pull/32464

[vapid pawn]

Hi @raven kernel I would like to apply for the Clawtributor badge! 🦞
PR Merged: #32119
Title: fix(plugins): allow hardlinks for bundled plugins (fixes #28175, #28404)
Merged on: March 2, 2026
Link: https://github.com/openclaw/openclaw/pull/32119

Summary: This PR fixed the "unsafe plugin manifest path" error that was breaking global pnpm and bun installations due to the strict hardlink rejection policy introduced in v2026.2.26. By allowing hardlinks for trusted bundled plugins, we restored functionality for package manager users while maintaining sandbox security for workspace plugins.

[dawn light]

hey, opened a fix for control tokens (NO_REPLY etc) leaking as visible messages to users — https://github.com/openclaw/openclaw/pull/32459 . would appreciate any eyes on it if anyone has a sec 🙏

[modern kestrel]

I know ACP is the new hotness... but I'm on Telegram with old and busted mode for coding-agent.

If you launch in foreground mode and tell it to ask for approvals when needed, it will... but in the last two weeks I've never once gotten one of those approvals with /approve or free form text to work.

I believe I have a fix for this now AND an improvement that actually tells you the /approve command syntax that will approve once.

But somebody please stop me if this actually works perfectly. The only way I've gotten tasks that need to access the internet in Codex is to tell coding-agent to use yolo.

I feel like I'm the only one using OpenClaw to code based on the issues I run into...

[vapid pawn]

@raven kernel good morning! I would like to apply for the Clawtributor badge! 🦞

• PR Merged: #32119 - fix(plugins): allow hardlinks for bundled plugins (fixes #28175, #28404) - Merged on: March 2, 2026

[dawn light]

Requesting Clawtributor role @raven kernel
https://github.com/openclaw/openclaw/pull/32367#issuecomment-3987988413

[icy plover]

I filed a bug, two PRs got opened that I went and verified didn’t fix it, I filed my own that is verified to fix it and today is got auto closed because one of the other PRs, verified not to fix it, was opened first.

I get that there’s a lot of PR noise but maybe the triage bot should also consider verification and which PRs are updated more recently? A bit of a frustrating experience trying to fix this bug and so I think I’m going to leave it alone and hope someone else actually fixes and verifies it. It’s a shame because the TUI interface seems to have several regressions and I’d like to be able to use it

[dawn light]

hey @raven kernel , requesting the clawtributor role. steipete took part of my PR https://github.com/openclaw/openclaw/pull/32367 and landed it on main with me as co-author, shipped in v2026.3.2 — https://github.com/openclaw/openclaw/commit/71cd3371372763e1f6c3d1fa7c11cb302285f619. got a few more PRs in the pipeline too

[burnt dagger]

Morning folks, what's needed to have https://github.com/openclaw/openclaw/pull/27275 reviewed and merged? I've done a lot of work and testing on this and I'm excited to contribute.

[viscid lotus]

Hi PR Reviewers/Maintainers, I have raised a very simple and small PR to add a locale support. Would appreciate if it could be reviewed/merged. Have updated the tests as well.
https://github.com/openclaw/openclaw/pull/30748

[tall prairie]

Link?

[dawn light]

hey @tall prairie, noticed that https://github.com/openclaw/openclaw/issues/22233 was closed pointing to https://github.com/openclaw/openclaw/pull/22270 as the fix, but that PR also got closed because the issue was marked completed. neither was actually merged — they just closed each other in a loop. the thinking block immutability auto-recovery fix never landed on main. worth reopening one of them to track it?

[icy plover]

I opened the but a first time that got closed by a PR https://github.com/openclaw/openclaw/issues/27674 without verification. So I opened https://github.com/openclaw/openclaw/issues/28180 and these two PRs were made: https://github.com/openclaw/openclaw/pull/28228, https://github.com/openclaw/openclaw/pull/28232

I pulled both locally and verified neither of them fixed the bug, so I opened https://github.com/openclaw/openclaw/pull/29117 which I actually verified does fix it, but it was closed in favour of 28232 (https://github.com/openclaw/openclaw/pull/29117#issuecomment-3987793746) which I had already commented doesn’t fix it https://github.com/openclaw/openclaw/pull/28232#issuecomment-3973465888.

maybe it does fix it now, I can try it again but at the time I made my PR neither did. I was worried they would get merged and close my bug again without proper verification of the fix - the tests aren enough someone needs to actually use the tui with tools and record it

[vapid pawn]

https://github.com/openclaw/openclaw/pull/33285

Ready to merge - fixes pnpm global install issue. All tests pass. Locally tested, CLEAN state, follows the exact pattern from approved PR #32119. Closes #29525, #29072, #21918, #32937. Thanks

[raven kernel]

there are conversations in the issue as of 2 hours ago.

[inland field]

I made a channel integration for instagram using https://github.com/supreme-gg-gg/instagram-cli can I just make a pr or is there anything else i should do

[jolly wolf]

keep it third party, we arent really adding more channels right now

you can get it added to #1474434870259224723 though

[inland field]

Coolio

[inland field]

Bet I’m working out the kinks rn after it’s ready I’ll just send the link here

[jolly wolf]

https://discord.com/channels/1456350064065904867/1474438720634748969

[marsh seal]

Any clawtributors with experience able to give me some advice? How frequently should I have my bot check and maintain this while I wait for human review? https://github.com/openclaw/openclaw/pull/29985

[digital frost]

Hello, looking for a review for a PR if someone has a moment:

PR: fix(security): prevent memory exhaustion in inline image decoding (#22325)
https://github.com/openclaw/openclaw/pull/22325

This prevents a potential memory-exhaustion/DoS issue in MS Teams inbound attachment parsing where oversized inline data:image/...;base64,... payloads could trigger big allocations during base64 decode, by adding a pre-allocation size estimate and basic base64 sanity checks, and enforcing both per-image and cumulative byte budgets throughout the parsing path.

Published test results within the comments as well

[sharp cargo]

just make sure it still merges cleanly. it could take weeks.

[spring yacht]

Hey! Looking for some help, our friends at AgentMail recommended posting here (crossposting from #clawhub) ... our skill clawdtalk-client is flagged as suspicious but it scans clean on VirusTotal (0/76). Would a mod be able to take a look? Details and evidence here: https://github.com/openclaw/clawhub/issues/568 .. thanks!

[unkempt jewel]

Can anyone get me the invite for iOS node app? Tysm

[dawn light]

Hey, opened a fix for the Codex toolChoice bug (#33421) — https://github.com/openclaw/openclaw/pull/33462. Happy to address any feedback!

[oak shard]

Hello, wanted to float this PR #23910 one last time for any maintainer interested in re-opening it. I saw that someone opened back up last week but it went stale again. Merge conflicts are fixed and tests pass now.

PR adds session:patch hook — simple and effective feature I have been using often.

PR: feat(hooks): add session:patch hook event to listen for session changes (https://github.com/openclaw/openclaw/pull/23910)

Discussion: https://github.com/openclaw/openclaw/discussions/24877

[dawn light]

Hey! Just opened https://github.com/openclaw/openclaw/pull/33500 fixing the Feishu duplicate bot-info fetch on startup. The single-account path wasn't passing the prefetched botOpenId to monitorSingleAccount, so it'd re-fetch every time. Small fix, added tests for the happy path, failed probe, and abort edge cases too.

[tall prairie]

Hooks are on the list. Keep it open

[timber pumice]

Hi! Opened PR fixing the bug of exec_approval empty list. Could someone with write access review or approve this when you get a chance? All checks are passing and WinnCook's review feedback has been addressed. Thanks https://github.com/openclaw/openclaw/pull/33346

[oak shard]

Fantastic thank you!

[marsh seal]

You think once per hour checks are overkill?

[sharp cargo]

probably, how often does one of those checks result in it actually needing care

i'd be doing it once a day at most

also like your pr should be small enough that for the most part the only merge conflict you have to deal with is the one where a different pr that addresses the same issue gets merged

[shut forge]

I just updated mine from main and now I have a test failure. Unfortunate, but better to find out this way than to merge the PR and then have breakage on main.

[sharp cargo]

take a quick look but usually those are transient

basically don't pay much attention unless it has an actual merge conflict

often main has breakage already

ok barnacle i guess

[deep tundra]

Hey Folks!
I know theres been a couple PRs and discussions on compaction models and I believe I've addressed all concerns about it being better to use the model that created the context that is getting compacted. But I came across two cases where we could add a bit more elegance to compaction failures.

1. Compaction fallback language, in cases where compaction is attempted with say a frontier model on a subscription but the compaction hits the usage capacity a fallback model is desired. https://github.com/openclaw/openclaw/discussions/33431 there is an associated PR https://github.com/openclaw/openclaw/pull/33396 that should meet requirements
2. In the cases that a frontier model with high thinking or just a really slow model is used there are cases in which a race is started between timeout and when the model returns it's response. When this happens it would be good to have the option to fail to "think = off" since it's not really a benefit for compaction in my understanding. If my understanding it could be changed to default to thinking is not changed and the feature is opt. Please see discussion and PR https://github.com/openclaw/openclaw/discussions/33430 and https://github.com/openclaw/openclaw/pull/33296/
   I'd appriciate a review when my turn comes up 😊

[sharp cargo]

starting to suspect that that change did not, in fact, obsolete my PR

[sharp cargo]

ah, i got scared by a refactor

[sharp cargo]

have a fix for a regression on llama.cpp backends: https://github.com/openclaw/openclaw/pull/33648

the fix is for https://github.com/openclaw/openclaw/issues/32916 which entirely breaks tool calling on qwen3.5 models (probably others but i didnt test those)

it's a one-liner!

[tall prairie]

Y2K was a one liner! (Probably)

[sharp cargo]

all i know is i could have been doing my actual job but i got sniped into fixing a regression

[tall prairie]

Job sounds way less fun then hacking on claw though

[sharp cargo]

i know right? i have to like interact with people instead of just turning rabid claudes loose on code

also i tried to make a video but 10mb is brutal

[hoary pendant]

I just pastied a couple of screenshots

[sharp cargo]

that probably would have been smart

anyway i'm enjoying my role as the brave defender of local model usability

and especially using a cloud model to do it

[hoary pendant]

I made that PR for per-topic agents you wanted @sharp cargo -- https://github.com/openclaw/openclaw/pull/33647

[sharp cargo]

hot

[hoary pendant]

fixing the codex/greptile things now

[sharp cargo]

i've considered vibing up a tool to basically vote on PRs by pulling them into your local branch and then running with them merged for n days

[keen spruce]

Am I misreading this completely, or is it currently telling the agent that it can generate up to ~6.5MB of text in its response when it's responding with that tool call? No wonder the sanity checks engaged. That many output tokens (>1M) are a bit above what's considered normal...

[sharp cargo]

Well it would have barfed on anything over 2k

but still

[hoary pendant]

I use quilt to cherry-pick PRs. It allows me to very easily rebase my local main with upstream/main and then test the patches for them applying, or just discarding the ones that no longer are needed or have a difficulty in applying. I love quilt!

[sharp cargo]

ooh neat

[deep tundra]

Ok, took longer than I expected but reopened both PRs and updated the discussion threads. So, ready to go when my ticket is called ;P

[sharp cargo]

ahhh i finally got my fix for tool result compaction working i think

yeah for the first time ever i've got it over 150k tokens in the context and not busting the cache to delete like 100 tokens from the beginning

[oak fiber]

I discover a TUI regression bug (#33866) and target how to solve it, then immediately submitted a PR (#33867) to fix it. However, many similar PRs were submitted shortly after :XD . I'd appreciate if someone could help review and merge it first. thx : https://github.com/openclaw/openclaw/pull/33867

[frail karma]

Looks like I've been posting into the wrong channel... forwarding here.

[sharp cargo]

thanks for totally rewriting my pr to actually be sustainable @untold pollen ❤️

[sharp cargo]

yesssss i think this is gonna be it, i'm gonna get this context all the way to the tool result compaction at 160k or wherever

ripppppp got to like 130k and then something super early changed

but thinkies are working!

[sharp cargo]

I am enraged.

slot update_slots: id  1 | task 82525 | old: ...  23: | 22:27 PST
slot update_slots: id  1 | task 82525 | new: ...  23: | 52:27 PST
slot update_slots: id  1 | task 82525 |      220      17      18      25      17      17      25      17      22   39630
slot update_slots: id  1 | task 82525 |      220      17      18      25      20      17      25      17      22   39630
slot update_slots: id  1 | task 82525 | n_past = 9659, slot.prompt.tokens.size() = 127716, seq_id = 1, pos_min = 127715, n_swa = 1```

Okay who put a timestamp in the system prompt again?!?!?!?!

There's also system events that are in the system prompt now instead of the conversation where they used to be???

I'm never going to get a single context to 160k tokens at this rate

https://github.com/openclaw/openclaw/commit/5b8f492a48fe2e1f467f78bfa06647cfc7c6f660

grumble

[keen spruce]

I hate that prompt code; it's so convoluted. But if I read that diff correctly, events are now added to the system prompt instead of the user prompt, invalidating the whole cache every time they happen? Is so, that's a MAJOR regression and can be very expensive for some users...

[sharp cargo]

Yep! I already have a patch which puts them back. There's an attempt at injection protection in that commit too which should make the cache bust unnecessary. I need to figure out how to write a test for cache busts ugh

I have like four branches soon to become PRs after I bake them on my agent

[keen spruce]

Yeah, my first instinct was to write a test case for system prompt stability...but I can't trigger unknown injection sources to inject inside the test case like I could do with the current time by just waiting 61 seconds...

[sharp cargo]

oh i also had a 30 minute timestamp change somewhere at token 9.7k

i hope that's just the system event thing

[keen spruce]

9.7k sounds like it could be the end of the system prompt. The prompt itself is now cached in memory, so it shouldn't change---only stuff people append to it can change

[sharp cargo]

i'm basically just manually trying to get a single context all the way from opening the session to the actual context window limit

and then having an agent chase down every time i have to recalculate my context for 20 minutes

the nice thing is that i get to interact with a 122b model when doing this so it actually does things by itself

[keen spruce]

My main session hasn't been reset for weeks...only compacted. But then, I am very careful with what versions I upgrade to...

[sharp cargo]

which also uhh exposed a fun bug with thinking!!

anyway i'm so close to validating the original concept which was fixing the tool result compaction pass to reclaim way more per pass and also be way more accurate in its estimations

i validated a fix for one bug and fixed like 3 others in the process

[keen spruce]

btw, do you remember which maintainer handled the last system prompt issue? I don't want to ping a random one... (and IMHO, this needs a ping)

[sharp cargo]

argh my fix for it was apparently ineffective i think

no wait i psyched myself out of course it invalidated the cache one more time

[rough flint]

Hey folks! 👋I'm a contributor here — I've got two small PRs open that address the undocumented commands.ownerAllowFrom gap (issue #25344). Ran into this myself setting up a self-hosted instance and figured others are hitting the same wall.

#25357 — one-file fix: adds a warning log when owner-only tools (cron, gateway) are silently stripped, so you actually know what's wrong instead of just seeing the agent can't find the tools
(https://github.com/openclaw/openclaw/pull/25357)

#25424 — matching docs: documents commands.ownerAllowFrom in the tools and pairing pages, since it's not mentioned anywhere currently
(https://github.com/openclaw/openclaw/pull/25424)

Both are small and verified against v2026.3.2.
Would love any feedback — happy to adjust if the approach needs changes!

[daring sable]

Hey Folks! 👋 I'm a contributer here

I've got a PR open for a new Kudosity SMS channel plugin: https://github.com/openclaw/openclaw/pull/32417

Adds outbound SMS via the Kudosity v2 API with config adapter, onboarding wizard, webhook parsing, and full test coverage. All automated review feedback has been addressed and CI is green (the only failure is the known flaky root-alias.test.ts Windows timeout that's also failing on main).

I'd appreciate a review when my turn comes up 😊

[plain perch]

Hi,
seems like I found a bug with openclaw onboard --install-daemon
Here is the PR :
https://github.com/openclaw/openclaw/pull/34306

[tall prairie]

The cache is busted again on main?

[proven moon]

Do providers return cache hit metadata? If nothing else I'd like to be able to catch issues with cache locally. For opus use this is rly important

[stiff flame]

Hey folks! 👋 I have a focused PR up for review: https://github.com/openclaw/openclaw/pull/15639
It fixes a race condition in local embedding initialization (prevents duplicate model loads / VRAM spikes) and includes test coverage for the concurrency path.
It’s a small, scoped change, and I’m happy to address feedback quickly.
Would really appreciate a review when it reaches your queue — thanks!
Would appreciate a quick look when you have a moment 🙏

[tall prairie]

Yes. /status should surface cache hit rates and there are a few dashboards that can also be used to see hit rates.

[rustic remnant]

With the overwhelming number of PRs how does one actually get picked for merging?

[white galleon]

small focused PR with a high quality fix

[rustic remnant]

I am following one that fixes a regression that is affecting me and it is size XS with experienced contributor so should get merged in smoothly? #33137

[rustic remnant]

Yay merged

[sharp cargo]

Yep I’ll push a pr later today I guess

It’s a “security fix” it’s very annoying

[tall prairie]

Thank you, one before the next release would be great

[sharp cargo]

Yep just always want to test stuff locally first 💜

[earnest smelt]

Hey fellas, I just joined this community, hope I can catchup with lots of happening in the OpenClaw space right now! 🙂

[sharp cargo]

https://github.com/openclaw/openclaw/pull/34794

Hope it passes checks, been running locally fine for a while

I have Claude working off and on on a tool to just detect cache busts more accurately but it’s a long way off.

I did manage to burn like half of my Claude max 5-hour allotment last night fixing like 3 other bugs in the process of validating that one feature-fix (tool compaction is much more tame now ..)

[carmine mural]

Hello there! 😄 I’ve submitted a PR to add MiniMax TTS support here: https://github.com/openclaw/openclaw/pull/27880

I’ve spent the last week addressing the automated comments and linting requirements, so the build is green. I’ve been using this branch locally to access my custom MiniMax voices, and it’s working great.

I’d highly appreciate a maintainer’s review whenever you have a moment! 🙏 Thanks!

[tall prairie]

Best tool is mitm proxy and watching the request/response it’s a pita though

[sharp cargo]

yeah i'm writing one of those but like you say, pita. Slot cache debugging on llamacpp is at least helping

[tall prairie]

No need to write one it already exists it’s called mitm proxy 🙂

[sharp cargo]

well sure but then scope creep happened and i wanted to also enrich the responses with caching data since llamacpp doesn't do that

@tall prairie automated review found some stuff, think i fixed it, but don't merge until greptile likes it 5/5 🙂

[sand sand]

hey folks, i've wanted to use openclaw for a while now. finally got myself some inference thru OpenCode Go. but it's not available as an onboard option. i raised an issue few days ago hoping that someone would pick it up, but it was not. so i raised a pr that closes the issue.

https://github.com/openclaw/openclaw/pull/34838

it might not be ready to merge. (idk this is my first time here). so please let me know how i can get it ready to merge.

• ci passes (1 fail. 404ing when downloading bun in the workflow)
• resolved greptile's review.

[tall prairie]

Have you tried the custom provider option?

Just curious

[sand sand]

no actually. i just looked to see if whatever provider i have was listed. i've used it via openrouter, google, openai codex. i was not aware of custom provider until someone tagged my issue in theirs.

[tall prairie]

We’re not against new providers just wanted to give you workaround

[vocal geyser]

I have a question last 2 weeks I've been working on tts. Initially I started by integrating qwen3-tts as a custom TTS provider but i found out that currently no new providers are considered.
So i completely scrapped that approach and instead went the correct path and just fully implement the OpenAI TTS API. still doing manual tests with verification videos is this something that will even be considered?

[sharp cargo]

lol codex is teaching me about features that i didn't know existed because i'm breaking them with my change. excellent

i gotta put up my other PRs so they can marinate i guess

i'm just like, are these too big? can i make it tinier?

[atomic nimbus]

Hey! Raised a PR to fix issue #29100 - Cost metadata support for auto-discovered Amazon Bedrock models. PR #31160: https://github.com/openclaw/openclaw/pull/31160 - Adds a costs map to bedrockDiscovery config so Bedrock IAM-auth users can finally see USD cost estimates in /status and /usage (not just token counts). Would love a review! All tests pass. Thanks! 🙏

[hoary pendant]

https://aws.amazon.com/blogs/aws/introducing-openclaw-on-amazon-lightsail-to-run-your-autonomous-private-ai-agents/

[red sparrow]

Could I get some thoughts on https://github.com/openclaw/openclaw/pull/34009 ?

[red sparrow]

Hey! I was thinking about adding support for the Perplexity Agent API (a /v1/responses based API to access Claude Opus, GPT-.2 etc)

From what I can see, endpoint compatibility here seems limited to /chat/completions and /messages. Does that mean a custom provider implementation would be needed (similar to how Grok is handled)?

[oak fiber]

it there a specific roadmap about how openclaw going to do next recently？

[muted frost]

I'm looking for a maintainer to help me add a channel to openclaw. it's like whatsapp but you don't need a phone number. called aliceandbot

[glacial lodge]

Hi folks, I just discovered that my PR for adding a model provider (StepFun) got closed. The reason is that it's already supported in OpenRouter. For this, I want to explain the difference:

Integrating platform.stepfun.com directly (via this PR) offers significant performance benefits for OpenClaw users—specifically, much faster response times and higher rate limits.

For context, OpenRouter strictly caps the daily number of requests depending on a user's credits, with a maximum limit of 1,000 daily requests. In contrast, the StepFun platform does not limit daily requests at all, and the limit is 1,000 requests per minute (RPM) even for a light user.

I believe this massive difference in capacity justifies the direct integration.

Any comments / suggestions are welcome. I do hope that maintainers can reconsider this PR.

https://github.com/openclaw/openclaw/pull/27211

OpenRouter API limits: https://openrouter.ai/docs/api/reference/limits
StepFun platform rate limits: https://platform.stepfun.ai/docs/en/pricing/details

And there is concurrency limit as well. That significantly affect the experience of OpenClaw use. I can provide proof to support the claim.

[sand sand]

okay. the custom provider seems to work. will run with it for a bit to see if nothing breaks.

what does this mean for the first-class support? i think it's still worth integrating?

[keen spruce]

In my opinion, for all that it's worth (which isn't that much here), everything that has a fully openai-compatible API already has first-class support. What we're doing for many of the providers is like putting USB ports with funny shapes on a computer that only take one kind of device, so the user has it "easier" to plug their new mouse in. "The Logitech mouse plugs into the triangular, red port, the Razor mouse plugs into the blue 5-pointed star, except when it has LEDs then it goes into the six-pointed cyan star, ..."

[glacial lodge]

Thanks for the clarification. I understand the maintainenance burden and have huge respect for what the maintainers have done to the project and the community. For providers that's not on the list, users would still face more friction there, like manually entering the base url and the complexity of <provider/model>. Such models may be supported equally from implementation perspective, but users of these models are not treated the same.

For my specific PR, my intention is to add the provider first and then will add coding plan and anthropic-compatible config later. Sort of incremental rather than an overwhelmingly big bundle.

With that said, I'm happy pivoting to a pure documentation update/change.

I'd suggest some more automatic / prompt ways to handle this provider config, instead of the hard-coded approach.

Appreciate the discussion on this matter.

[final ocean]

Like what providers? What do you suggest?

[final ocean]

This

[keen spruce]

Maybe we can establish some kind of standard to package all the needed data into some oblique string that the providers can publish to simply cut&paste in? There are quite a few variations on the API you want to configure, including the prices. Make it QR-code-payload compatible, and users could scan it with the app.

[final ocean]

Mcp had this at the start. A simple “vllm” configuration string is VERY doable. Can easily work like openclaw provider setup <string>

it will then preconfigure the custom model inference using openai spec

Ollama also have /models endpoint for discovery etc. So easier to add custom model configs into config and it works out of the box

If people want a “smoother” experience

[glacial lodge]

@final ocean the context is here (provider and PR). My suggestion is to have some built-in intelligence to let people configure using natural language or QR-code as Henry suggested (even better).

[glacial lodge]

Or somehow decouple the model / provider selection from the main code, like hosting in a separate catalog repo (and publishing manifest - package maybe) - this means any change to this separate repo is a data entry PR (not a core OpenClaw code PR). OpenClaw will keep a small resolver that loads a pinned snapshot, and optionally use catalog update to fetch newer versions.

[final ocean]

Too hard basket. Why do you need “intelligence” when its just a config??

[final ocean]

Also problematic as models chabge too fast. LiteLLM and openrouter do this and it trickles down to pi then into openclaw

[keen spruce]

naturl language leads nowhere with most of the datafields. "supports dev role"? "suuports system role"? "name of the token field"? most people can't even repeat those phrases, much less give them in natural language

[final ocean]

If you want an easier way to configure/override sure. Not sure what the ask is

I think small labs just want to have thier models show up better. I hear you, but we cant maintain code and spec for what is already supported

[keen spruce]

I'll think about a data format tomorrow, I have some ideas on how to make this simple (K.I.S.S. principle!), but I'm about 3 hours late for bed, so I don't trust my own brain atm ;)

[final ocean]

I dont “monitor” this channel so dm me when you have something

[keen spruce]

will do

[glacial lodge]

I didn't think through when I said natural language. Right. It's not deterministic. but hard-coding model config during onboard isn't ideal, and I see catalog/manifest maintained separately a viable direction.

[sharp ivy]

Hi! Contributor-in-training here. I have a small stack (6) of PRs. Half of them are miscelleneous fixes. The other half are cautious (I think) attempts to extend the hook system in line with existing patterns, and with an eye to open discussions, issues, and other PRs on this core topic. The CONTRIBUTING.md suggested opening a discussion, but it got closed. My goal with the hooks is to enable richer security plugins, one of which I have developed and have been using extensively for about a month. Is this the right place to find humans to talk about this with? If not, please let me know where to go. Thank you!

[oak fiber]

agreed

[glacial lodge]

I guess under the current setup, to some degree, model providers who got added earlier have some sort of advantage, and it's not fair for new model providers / router services. Most of the models already on the list are either openai-compatible or anthropic-compatible, I guess we would not remove them?

And if the standard is whether a model is already supported on OpenRouter, almost all models are now available on OpenRouter. OpenRouter has competitors too, and they would never have the chance of showing up?

[frail karma]

I think there's a gap to formalize the license definition for OpenClaw skills.

I'd like to propose that we add a license definition feature for skills! Detailed PR created for this (with screenshots)
https://github.com/openclaw/clawhub/pull/603

Feedback appreciated! 🙏

[lunar goblet]

Hi @keen spruce, I am a co-founder of StepFun (http://yibozhu.com/). I agree 100% that USB ports should be standard cross mouse brands. The problem is that there are already many "USB ports" for "Logitech", "Razor" etc. with different fancy shapes. Using it as a reason to reject new providers is in a way suppressing new mouse brands. I always believe that an open community should treat new contributors equally, if not more friendly, as/than the existing players. More choices and more competition is generally good for users.

I apologize if @glacial lodge offended or bothered you. It's a bit frustrating for us since our PR has been hanging for almost a month (dated back to the earliest one https://github.com/openclaw/openclaw/pull/10857). While we see the usage of our model gradually climbs to #1 daily on OR (https://openrouter.ai/apps?url=https%3A%2F%2Fopenclaw.ai%2F), we are still rejected by OpenClaw for being shown as an independent provider. We are a bit confused of what more we can do to get accepted.

We look forward to your new and fair solution!

[bold canopy]

Hey team! 👋

I've added SGLang (https://github.com/sgl-project/sglang ) provider support to OpenClaw. Ready for review. https://github.com/openclaw/openclaw/pull/35572

What it does:
• Auto-discovers models from SGLang server (port 30000)
• Adds SGLang to provider selection UI
• set SGLANG_API_KEY and it's ready to go

Changes:
• Core provider logic + auto-discovery
• UI integration (appears after vLLM in provider list)
• Setup prompts for base URL, API key, model

Let me know if you have any questions!

[cosmic copper]

Hi mates , respectful bump on my PR: https://github.com/openclaw/openclaw/pull/29075

I understand the PR queue is very large, so I’m not expecting fast turnaround.
I’m posting a concise “what I tested” to help triage:

What I tested:

• Live manual Zalo checks (real accounts): text, link with/without preview, image, sticker, unsupported payloads.
• Verified unsupported payload fallback logging + DM policy enforcement.
• Ran: pnpm vitest extensions/zalo/src --run
• Ran: pnpm vitest src/infra/safe-open-sync.test.ts --run
• Ran full local pnpm test (passed)

If this is easier to merge in smaller pieces, I can split the PR immediately.
A quick directional comment (keep as-is / split / close) would be enough and I’ll follow it. Thanks.

I’ve noticed a few recent smaller PRs touching the same areas this PR already addresses.
Reviewing/splitting this one now could reduce duplicate follow-up PRs and keep the queue smaller.

Have a nice day 🤙

[modern kestrel]

A little PR for fellow clawtributors working on the gateway: stop restarting the gateway when .test.ts files change:

https://github.com/openclaw/openclaw/pull/36211

[final ocean]

anything specific to sglang you cant do with standard custom providers?

[carmine mural]

Would it be possible to get a code review on this PR? I can see that the workflow completed successfully 🤩

[silent bolt]

Hi maintainers! I've submitted a PR to add Prismer as a community channel plugin (#36250) and opened a discussion about potentially integrating it as a built-in extension (#36264).
Prismer is an A2A messaging network — 80+ agents in production. The OpenClaw plugin implements the full ChannelPlugin interface: outbound, WebSocket gateway, agent discovery, and two bonus agent tools (web context + document OCR).
Everything is published and tested:
• npm: @prismer/openclaw-channel
• Source: https://github.com/Prismer-AI/Prismer
• TS type-checked against plugin-sdk — zero errors
Would love guidance on what's needed to move toward a built-in extension under extensions/prismer/. Happy to add vitest tests, adapt to monorepo conventions, or adjust anything per your feedback.
PR: https://github.com/openclaw/openclaw/pull/36250
Discussion: https://github.com/openclaw/openclaw/discussions/36264

Hi! I'd like to contribute a new channel extension — Prismer (A2A agent messaging). Already published as a community plugin (@prismer/openclaw-channel) with a PR for the docs listing (#36250).

Before I submit a PR to add it under extensions/prismer/, wanted to check: is a Discussion (#36264) sufficient, or should I get explicit approval here first?

[zealous plover]

Does it unlock that the existing community plugin can’t already do or just discovery/distribution?

[silent bolt]

Think of it this way: other channels are walkie-talkies — agents shout raw text at each other. Prismer is a shared library with a phone system — agents publish knowledge to the shelf, then just tell each other "check page 42."

[jolly wolf]

s!warn @silent bolt AI slopping long paragraphs is not appreciated here, use your human voice and answer questions and send messages yourself

[silent bolt]

i just provide an accurate feature specification.

[jolly wolf]

No, you're spamming the channel with slop.

[tall prairie]

i suggest taking it to #1457803836877176914 @silent bolt

[sonic ermine]

i would rethink this a lot , it does make a lot of sense to support native frontier labs and inference providers because the cost is much smaller and the latency is much smaller... some of these have been around for much longer than openrouter ... talk about wierd internet money , well there's one example of it , versus straight up model labs hosting their models for you at 1/10th of the price with actual caching ...

my point is , we should consider rethinking this approach , especially since the labs have folks that are contributing the code that needs to be maintained

[serene nimbus]

Hey guys, where can I go for support, have a few questions

[final ocean]

@sonic ermine not disagreeing with you, open to solutions that dont complicate the problem future and works for all

[serene nimbus]

Installing openclaw on windows and need some help

[final ocean]

#1459642797895319552

[sonic ermine]

well this is super encouraging , because i've literally recieved dms from the representatives of a bunch of labs last three weeks , if i clearly mark the PRs from legit reps and legit labs , would that help ?

i mean i got like two DMs , but i see more wallowing in the queu (and they're legit)

[bold canopy]

Thank you for the feedback. Why it's needed and specific:

1. Auto-discovery queries "/V1/models' automatically (no manual model config) 2. Onboarding UX (Ease-to-use). 3. Growing adoption SGLang is increasingly popular for high-perf local inference. We already have vLLM, Ollama; SGLang fits the same pattern.

[zealous plover]

imo I think the custom providers is still a good fit, but maybe you can open a GH discussion for this?
I've seen a lot of similar requests and unless there's a large user demand, the answer is usually keep it as a custom provider. (But maybe start w/gh discussion & we'll see what others have to say)

[final ocean]

Lets move to DM cc @jolly wolf

Yea will work. I need to hear from providers directly so we can help address thier feedback and see if we can work something out

[final ocean]

Feel free to start one

[frank berry]

Hey Team - probably for the maintainers, but I got invited to https://github.com/OpenClawHQ a few hours ago. It looks like probably a non-affiliated repo but could easily confuse people.

I'm just ignoring it but thought I'd call it out as it gives me weird security vibes.

[frail karma]

The Skills license feature is now ready and waiting for a human to review and discuss. It's meant to be a step towards introducing licenses into OpenClaw's skills workflow, much like Google / Getty Image search for commercial image assets.

The AI bot reviews are all completed... took a bit of time to nail this but finally did it!

• GreptileAI summary is now 5 out of 5
• Codex Agent gave a 👍
• Vercel Agent Review successful

[zealous plover]

That user on GH is also spamming the comments to promote his "OpenClawHQ" via comments & PR's

Comments on other's PR's:
https://github.com/openclaw/openclaw/issues?q=commenter%3AJah-yee OpenClawHQ

All of these PR's in the list have spam comments promoting his Org/Repo

[modern kestrel]

I've got Telegram ACP Topics (threads) binding working in this PR, also fixed some corner cases that made it impossible for me to actually spawn ACP at all (-- turning into emdash on Mac Telegram client).

If someone using Discord ACP could just confirm that things are working for them still that would be a big help!

I'm also working on something that I think is better... might be able to work some of it into the ACP approach though too.

https://github.com/openclaw/openclaw/pull/36683

[final ocean]

@versed kindle

[versed kindle]

I already built this functionality yesterday? not released yet. is this related? have you checked it out?
https://github.com/openclaw/openclaw/pull/34873

[modern kestrel]

I was told last night it landed on main… I saw the commit on main… tried it, but it didn’t work 😂

Did you fix all the same issues? There were several.

To clarify: I tried what is on main. Haven’t looked at the PR link you sent and whether that was merged yet or not.

[versed kindle]

ah got it. I checked the PR out, it looks good to me, but I haven't tested

[modern kestrel]

Oh yeah that’s is the PR I’ve been trying. It’s already on main.

The first screenshot in my PR shows that topic binding in TG doesn’t work with ACP. Some issues are just annoyances like the client converting double dashes to an emdash, but once that’s out of the way it still says binding is not supported in TG. My PR fixes that.

[versed kindle]

@modern kestrel i can ask the agent but while you are in the context, did i mix up telegram/discord code, and put up channel-specific code where it shouldn't be?

[modern kestrel]

I’ve got to make dinner for the kids for about an hour but will post a video review after on my PR. I believe I’ve tested all the corner cases.

[versed kindle]

I'll also give it a run and merge it if it doesn't break anything

[modern kestrel]

I’m not positive? Maybe you missed some commits? The docs on /focus an /unfocus were still discord specific as were the docs on spawn saying only discord supports threads. I saw nothing on main that added telegram support. Maybe you did a lot of the plumbing to make topics work but just missed a final step or lost a commit?

[versed kindle]

ok will look again

I agree that telegram should support the config values and the slash commands you mentioned

what I'm not sure still is: did you set telegram up by adding config values in bindings and such, or did you purely try to use slash commands?

@final ocean thank you btw for bumping

@modern kestrel any chance you might add me as collaborator on your fork? then I can push to the same branch without having to create a new pr

username: osolmaz

[modern kestrel]

Yes will do in a minute.

[modern kestrel]

I just tried to figure out how to use ACP from what crumbs I could find. All I knew was the slash commands 😂

[versed kindle]

ahh just created a new PR 😅

if you don't have time, no need to worry, I will credit you while merging

[modern kestrel]

Just added you to my fork

You can DM me if you want. I can pair on this too if you want...

[versed kindle]

So there will likely be a release tonight, and I'm ok with polishing telegram experience more before we do that, it's gonna be a big release

I am ok with pairing as well, I'll just sleep in a bit

also, you should check out the doc changes in my pr. it tells how to set up topics permanently with codex claude code etc. that was the flow I tested, so I'm not surprised slash commands weren't working

[modern kestrel]

Ah... gotcha. Yeah I didn't read the PR I just dove into trying it with the /acp command that I knew existed 😂

[versed kindle]

ah sorry, btw can you also add dutifulbob as collaborator? forgot that I was using my clanker 😅

I don't wanna discredit you

[modern kestrel]

Done

[magic vine]

Probably should have put a configurable archive duration on the original discord autoThread PR, but better late than never 🙂 https://github.com/openclaw/openclaw/pull/35065

[modern kestrel]

@versed kindle - Do you know, offhand, why ACP is only allow "yolo" or "read only" or "never" for approval? I believe that can be changed and I think I had it changed on one of my branches last night... just wondered if there was a reason why it wasn't set to be interactive. I have exec approvals showing buttons in Telegram for old-style /coding_agent invokes now and my alternative-to-ACP approach is using button approvals as well (relaying what Codex is showing)... so I was curious why ACP wasn't exposing the "ask me" mode. I can add that if the ACP library supports it.

[versed kindle]

hmm i'm not sure about these chabges after this

@modern kestrel is it possible to split this pr? I'm not sure about the spawning and slash command interface stuff. But any Telegram-related fixes we should still ship

[modern kestrel]

I think that it's saying it's parity with Discord? I didn't invent the ephemeral-ness unless I made a mistake and persisted the config when they should be ephemeral... I can fix that.

/acp spawn codex --mode persistent --thread here --cwd /Users/huntharo/github/jeerreview --label jeerreview4

My tests were always like that... I was telling it to make them persistent...

[sharp ivy]

Hi all! Small stack of PRs ready for review. Goal: expand Openclaw's hook system carefully, misc fixes. 5/5, green aside from CI flakes.

Hooks series: https://github.com/openclaw/openclaw/issues/36671

Independent smaller fixes:
https://github.com/openclaw/openclaw/pull/30310 — channels: default account in multi-account configs
https://github.com/openclaw/openclaw/pull/30323 — browser: include current page URL in tab-targeting responses
https://github.com/openclaw/openclaw/pull/33845 — slack: persist thread participation cache across restarts

Enjoy! 🧁

[magic vine]

I had a question about ACP: Any interest in routing the output to wake the agent/session? I thought it was going to be a drop in replacement for how I have OC run/supervise claude code and codex via tmux, but this doesn't work because the agent can't see the responses from cc/codex.

[versed kindle]

So I would say you can set up your config using those binding arguments for now. And then we can polish the slash command binding in the next iteration. What do you think?

[versed kindle]

For the codex session to be persistent it needs to be saved to config but you weren't doing that no?

[modern kestrel]

I think if the release notes mention Telegram Topics support for ACP that it's going to result in a slew of support questions if the slash commands don't support it.

But if it's left out of the release notes or carefully mentions config-driven only... might be ok

[modern kestrel]

I don't see them in my Codex threads list no.

[magic vine]

Just realized I never got my clawdributor role, @jolly wolf could I get that whenever you get a chance? https://github.com/openclaw/openclaw/pull/800

[red sparrow]

I am surprised that flow^ isn't automated

[jolly wolf]

It’s intentional, because of the amount of people who try to sneak in malicious PRs

Even if we banned them, technically they are still a contributor to the repo forever, and I didn’t wanna have dozens and dozens of checks for bans and deletes and stuff, easier to just do it manually still

[versed kindle]

Telegram doesn't have an ephemeral thread option. Telegram topics are equivalent to Discord channels. I should probably add clarification that for persistent use of codex, it needs to be added to config.

I'm using voice typing, by the way. Sorry if it's annoying

[magic vine]

Thanks!

[red sparrow]

technically they are still a contributor to the repo forever

Never use the contributor list as the source of truth!! You could have a bot maintain its own allowlist. Something like a /claim-role command that triggers github oauth, then the bot verifies with is:pr is:merged author:{username} repo:openclaw/openclaw, meaning you get a list you can actually manage from the bot's collected data

And in the event of a revert, that's very easily traceable so you can auto-remove the role or take action against the user! With 1k+ contributors the manual flow doesn't seem viable long term IMO

For context I used to be senior mod on the OpenAI discord (scaled to 3.5M+ members) and that server had a bunch of internal automations for similar validation flows

[jolly wolf]

Yep, that’s exactly the list of checks I said I didn’t wanna do 😂

It may be worth looking into later once we have the foundation and more official ways to track things like that, so that we could scale it to sponsors and employees and stuff too

We’ve had a couple situations so far of people stealing other people’s PR’s and getting them merged, and then later on we just change the attribution in the contributor list of the original author if it’s something that we missed. That’s the big situation that is really hard to account for

The edge cases are daunting at the scale of slop that we’re getting

[red sparrow]

lol I don't think it needs to exist tomorrow but with the fastest growing repo and 120k+ members it may be worth considering down the line

[red sparrow]

sure but you could make the process less manual eg /reassign-pr #34009 @kesku

[oak fiber]

It's really hard to contribute now. There are so many pull requests. When an issue is raised, if it's relatively easy to fix, several PRs will be submitted within five minutes. I once filed an issue and was the first to solve it, but in the end, my PR wasn't merged. I felt a bit frustrated 😟 But I also understand that there are so many open PRs right now—it's really tough.

[sharp cargo]

enlightened solarized enjoyer

[oak fiber]

I've opened a PR 2 days ago, to fix a security gap in the doctor command https://github.com/openclaw/openclaw/pull/34244
unlike other my trivial PRs, this pr is worth to take review i think

[lethal tulip]

Hey all 👋 I've been building a personal exec agent and
implemented something called AgentGate , a credential broker
pattern where the LLM only ever holds an agent-key, never
OAuth tokens directly.

Thinking it could be a useful extension for OpenClaw,
especially for users running cloud LLMs who don't want
their Gmail/Calendar tokens touching external servers.

Posted a discussion on GitHub here: https://github.com/openclaw/openclaw/discussions/36963

Would love feedback on whether this fits the extension
architecture before writing code. Happy to contribute!

[modern kestrel]

Those PRs are usually, in my experience, not good. I tested some of them just to make sure. Maybe that’s not true anymore as it has been a few days.

But I stopped filing issues if I intend to fix something myself.

[sharp ivy]

A hint I got from Krill, which I liked a lot, was: if you want to contribute specifically, write the PRs and THEN write an issue that demonstrates the problem and the solution at the same time. I can't vouch for how well the outcomes match the promise, but, it seems like a reasonable strategy to me.

[oak fiber]

yeah, that's exactly what I do now. If I find an issue n can fix it myself, just submit a PR directly without filing an issue. Otherwise, there will be several similar PRs right away, that's unnecessary.

[zealous plover]

#self-promotion plz - this just isn’t the channel for that

[keen spruce]

That's clerly a promo. You may not be selling something, but still...

[sour sapphire]

Hey everyone 👋

I've been digging into a long-standing LINE channel bug — requireMention: true in group/room configs has no observable effect; the bot replies to every message regardless. The root cause turned out to be three separate missing layers:

1. No mention gating in the inbound flow
2. No pending history for unmentioned messages
3. LINE absent from the DOCKS registry

I put together a fix covering all three in a single PR:

🔧 PR #35847 — Fix/Complete LINE requireMention gating behavior
https://github.com/openclaw/openclaw/pull/35847

📋 Issue #34438 (full root cause analysis + design rationale)
https://github.com/openclaw/openclaw/issues/34438

The fix brings LINE's group gating behavior in line with WhatsApp, Telegram, Discord, Slack, Signal, and iMessage — unmentioned messages are stored as pending history context rather than dropped or processed unconditionally.

If you have a moment, a review would be very appreciated! Happy to discuss the approach or answer any questions. 🙏

[lime crane]

Quick question, is it intended that is discord the only supported channel for ACP? Is it a planed feature to allow for ACP without one shot outside of threads?

As far as I can tell on a channel that doesn’t have thread support you can’t use a session that keeps context in a codex acp.

@modern kestrel Any insight you can give would also be appreciated

[versed kindle]

what you are asking is currently on main

• acp in discord threads
• acp in discord channels (persistent)
• acp in telegram topics

additionally @modern kestrel did some work that acp related slash commands work on telegram yesterday

it should be released hopefully today

[lime crane]

Thanks for confirming, I don’t currently use the discord channel. What’s stopping claw from just starting a context persistent acp session in other channels other than discord and telegram? Seems to be a hard coded limitation

[versed kindle]

now that we support persistent bindings, yes, we could enable for others. but the goal was to test each channel before releasing to make sure it works. releasing isn't as seamless and acp stuff need to be specialized for the idiosyncracies of each messaging app

which others did you have in mind?

[crisp linden]

Wanted to ask if anyone could update the pi-ai related dependencies in the repository - there seem to be some key fixes in there since the pinned version :)

[lime crane]

That makes sense, I think the web ui is an easy no brainer as you have full control over it, however I’m not sure how many claw users actually use it so might be best to focus on the more popular channels.

I feel implementing it into the popular channels like WhatsApp/Signal would be a good next move. What makes it difficult to just have some mechanism (similar to native claw subagents) to repeatedly interact with the same acp session in the background

I don’t understand why backend functionality is tied to the front end interface, although I have not properly looked at the inner workings of acpx

[versed kindle]

can you maybe elaborate the exact UX you have in mind? whatsapp community with subgroups? or using slash commands to focus in and ouf codex in the same chat?

[lime crane]

Either would work, but I think slash commands would be the most versatile. If slash commands were implemented could, a subagent also easily use the same underlying interface to acpx to read output and send tasks to the acp codex session?

The use case of manually entering the acp session with a slash command shouldn’t be the main focus imo (but is nice to have). I think allowing an agent to fully control the session is more important for me anyways.

[versed kindle]

I didn't get exactly

sounds like it describes something that is already possible, except for whatsapp/signal

did you try to build on main?

[modern kestrel]

If anyone with a Discord setup wants to test this to see if anything changed on Exec approvals (sample prompt that invokes npm with approval in the PR), I would greatly appreciate it:

https://github.com/openclaw/openclaw/pull/37233

I've never been able to get exec approvals to work on Telegram until this PR - the codes were always rejected. This presents the approval choice as buttons, clears the buttons after click, fixes the issue with the approval not working, and adds a (need review) system event heartbeat routing back to the requesting topic that allows exec to actually display output after the approval.

[modern kestrel]

The event waking changes can be dropped if too risky... they are isolated commits.

[magic vine]

@versed kindle I had a question about ACP: Any interest in routing the output to wake the agent/session instead of direct to user? I thought it was going to be a drop in replacement for how I have OC run/supervise claude code and codex via tmux, but this doesn't work because the agent can't see the responses from cc/codex.

[jolly wolf]

https://x.com/4shadowed/status/2029976890820178309
and yall wonder why we struggle so much :P

Shadow (@4shadowed)

@openclaw (That was 560 total open PRs by that one author)

X · Mar 6, 2026, 5:46 PM

[magic vine]

@jolly wolf just saw that you closed this PR (not mine). Was it because it was stale/triage, or just not interested in that fix? Until recently my agent in Discord could edit openclaw.json and message me once the gateway was restarted, but now they just stop. https://github.com/openclaw/openclaw/pull/29255

Not sure if I should open a PR to fix or just leave it alone

[jolly wolf]

That user is the one referred to in my latest tweet (also sent in #shadow-says-stuff)

im banning a lot of users right now for directly copying peoples PRs and spamming us

[wise helm]

oh makes sense

[jolly wolf]

there's likely another PR open that is the exact same fix, since most of that users PRs were dupes

[wise helm]

i created a pr and he/she/it commented and created another pr with fixes.

https://github.com/openclaw/openclaw/issues/37832

then saw sid-qin had 80 merged and 400 open and thought must be a maintainer

[jolly wolf]

maintainers will both have the orange role here and have a yellow maintainer label on their PRs

[wise helm]

my bad ✌️

[jolly wolf]

all good

[magic vine]

I'll keep looking, but I'm not finding one

[jolly wolf]

you're welcome to open one then!

[inland field]

@jolly wolf I dmmed u my instagram channel plugin could u pls go through it and tell me if there’s anything else to do

[jolly wolf]

shit

i forgot

[inland field]

Np lmao

[crisp linden]

Could a Maintainer maybe take care of this?
I could make a PR but I feel like that's kinda unnecessary here

[wise helm]

@jolly wolf

this guy used cursor for opening a draft pr https://github.com/openclaw/openclaw/pull/37981
based on my issue that i already created pr for
https://github.com/bmendonca3/openclaw/commit/c38f1528acaa48a91963f128a172f7005da4db5b

seems like same pattern

[jolly wolf]

it'll be in the next release

[crisp linden]

Thank you!

[jolly wolf]

they're in my queue for banned users!

actually they're already banned just havent closed their prs yet

[red sparrow]

soft bump! https://github.com/openclaw/openclaw/pull/34009

[final ocean]

i will review

ahh merged lol

[red sparrow]

appreciate the offer anyway haha, thanks for all your incredible work!

[tall prairie]

Good pr idea. We need to do that for embedding providers too.

[modern kestrel]

I had that user on my naughty list... I was surprised to keep seeing commits merged for them... but if you throw 800 PRs at the wall, some will get through right?

[wise helm]

can someone check this out? i spent two days finding this bug

https://github.com/openclaw/openclaw/pull/38322

[pure saffron]

Hi all 👋 . New contributor here. I noticed a few hours ago that main has broken CI. Here's a small fix: https://github.com/openclaw/openclaw/pull/38267

[oblique eagle]

I submitted an important PR to help separate data plane and control plane over two,weeks ago. Here’s an email I just received and my response:

“Closing this PR because the author has more than 10 active PRs in this repo. Please reduce the active PR queue and reopen or resubmit once it is back under the limit. You can close your own PRs to get back under the limit.”

This is perhaps the stupidest thing I have ever read.

What this has convinced me to do is cease any efforts to improve OpenClaw. Congratulations.

[sharp cargo]

Ok well sorry that you can’t handle contributing in the way they want you to

Like there are 5k open PRs lol

[oblique eagle]

That PR is about 3 weeks old at this point. I submitted about 17 in total, mostly to do with security concerns. I am not responsible for those who are abusing PRs, nor was this an expectation 3 weeks ago plus. But hey, not my house, not my rules. I’m just letting the maintainers know that they are throwing out babies with the bath water.

[jolly wolf]

https://github.com/openclaw/openclaw/issues/38283

i closed literally 2710 prs that were spam/slop, some people had upwards of 400 prs open.

blame the bad actors ruining it for everyone else, its unsustainable for us to keep going like we were

[oblique eagle]

And yet no consideration was given to the PRs sitting around for weeks…. Or the content of the PRs… Or their provenance. I get the attendant problems, but given how little attention the PRa have received since submission, to ask me to decide which are unimportant enough to close so that the others might be merged is a bridge too far. It’s not like the PRs were evaluated for importance or,usefulness; one was just chosen…arbitrarily? Given its security impact, it cannot have been chosen due to lack of merit.

[jolly wolf]

given how little attention the PRa have received since submission
we constantly are reviewing prs all the time, and we provide this channel for people to help discuss their issues as well.
the only criteria used for bringing the limit to 10 on existing prs was time, the first 10 prs were kept. anything more would have been unscalable or take days of work.

and like barnacle said, if you have other prs you think should be open instead, you can close your own and reopen the ones to keep

[oblique eagle]

My entreaties in this channel to evaluate important PRs went unnoticed or (certainly) not acted on. Again, I don’t take issue with the problems you’re facing, just the after effects of the chosen solution. But in any event, I won’t waste any more of your time.

[tall prairie]

Cutting the amount of open PRs helps us get to legitimate PRs. GitHub barely has any moderation tools so we have to work with a blunt cleaver.

[last frigate]

It's the right decision.

[modern kestrel]

If we did not (I'm going to say it here anyway for people reading from the sidelines) give guidance on how to manage max of 10 open PRs:

1. Nothing stops devs from making a PR against your own fork back to main on the fork - I do this all the time - granted, it won't probably do any CI, but you can run code reviews in Codex, run all the builds locally, and be 100% confident that a PR is "ready" to take 1 of the 10 slots and only promote them to the openclaw repo when they reach that state
2. Draft PRs - Did we suggest to people that they can make Draft PRs or did we include those in the cap? I'd probably include them in the cap... they can be just as distracting... if someone wants to make a Draft PR... make it on their own fork instead.

[jolly wolf]

drafts are included, yeah

no one except maintainers have really been doing draft prs though

[modern kestrel]

Oh no... trusted contributor went away 😭

[final ocean]

STOP

We dont need peoppe to fix tests for CI

Maintainers will manage this

Sorry if im touchy about this 😂🫡

[wise helm]

@lusty vigil could you take a look? This touches gateway send context propagation and agent/session-key handling.

[tall prairie]

Maybe we can figure something out but a raw count didn’t make sense

[oak fiber]

well, i have 4 opened PRs which not hit the limits of 10 opened PRs, wishing a review to be quickly

[oak shard]

Hello, I might be a little lost, but I no longer see the pr-thunderdome-dangerzone channel and I am wondering if I broke a rule on accident? When I first joined I forwarded a PR from here to there assuming that was where it should have gone. Is this why I cannot access it?

[keen spruce]

No worries, that channel go retired

[oak shard]

ah I see thanks

@tall prairie if https://github.com/openclaw/openclaw/pull/23910 is still relevant do you mind opening it back up to the queue? 🙏

[lime anvil]

totally reasonable change imo as a fellow open source maintainer. Only so much stuff you can review at once

[oak shard]

This should be resolved. It likely happened when I renamed the branch and then reopened the PR.

[hoary pendant]

Is there a channel in this discord to discuss qmd dev? Should it be a plugin or channel sub thread?

[final ocean]

@hoary pendant technically its a plugin

[silk kestrel]

Hi! Could anyone review this PR?
PR #34205 https://github.com/openclaw/openclaw/pull/34205

It adds result-modifying hooks for tool calls. This opens up a few useful scenarios:

• General testing / wrapper plugins: wrap or mock tool results without modifying tool logic
• Input guardrails: intercept and sanitize tool results before they reach the agent, blocking untrusted content that could be used to attack the OpenClaw agent
• PII / confidential data sanitization: scrub sensitive data from tool results before they're seen by the model and OpenClaw.

All tests pass. The only failing CI check (secrets) is unrelated to this PR. Would appreciate a review!

[polar schooner]

This one was "approved" by arosstale but then marked stale and closed. It's still really important and relevant. Posting here as #pr-thunderdome-dangerzone seems to be gone. https://github.com/openclaw/openclaw/pull/22917

[keen spruce]

Sorry to hijack this, but as you (presumably) just had your finger in the tool call code, you might have some insight on a random idea I had earlier: How easy/hard/feasible would it be to PR a very simple tool into openclaw: "custom" with an open list of parameters. When used, openclaw would call an executable configured in the config, with those parameters 1:1 (and the agent/session id). That's it, no semantics, just a simple callout that can work as a replacement for exec---for anything a user might want their agents to do without giving them full shell access. A very simple extension hook for people who don't want to dive deep into their openclaw but have some idea how to write a shellscript.

[silk kestrel]

Acustom or custom_script tool sounds like a great addition, something easier than building new plugins.

The tool can manage a list of scripts and supports commands like add, delete, modify, list, and run. Using it, the agent can list available scripts and run the script the user requested.

I haven't checked the default tool implementations, but I don't think that would be particularly difficult to add. It might add some RCE risk (if the attacker can overwrite the script) but that's nothing new for openclaw. 🙂

[keen spruce]

I was thinking even simpler. Just a single fixed entry, and then you can branch out in there based on the first parameter. Feed the info on what is available there as skill to your agent.

For overwriting I have two complementary ideas: (a) Put a warning into the doc that the file should not be writable by the agent, and (b) have the call be hardcoded to "<configured_executable> -h --help --agent <agent> --session <sesion> -- <parameter list>" The "-h --help --" would trip up nearly every existing executable, so setting the config value and pointing it somewhere would be useless. The agent would need to write a custom script and chmod+x it (sorry Windows, going by file endings bites you here) in addition to writing that config value to use this to break out.

huh, would a config param for the file-edit tool "rejectEditingExecutableFiles" be a useful thing?

[silk kestrel]

In general, i think it would be good to have specific directories or files configured as read-only so that the Openclaw agent cannot modify them using write/edit tools. If it marks the scripts as read-only, it could prevent accidentally or unintentionally overwriting them. Does openclaw already supports such a config?

[keen spruce]

I don't think the file editing tool has any restriction except for a "workspace only" one that can be enabled. In general, when exec is enabled, any such restriction would be useless. But when it is off, editing existing shell scripts is a hole.

[wise helm]

Hello maintainers,

could you review https://github.com/openclaw/openclaw/pull/38322 (issue https://github.com/openclaw/openclaw/issues/37832)?

In sandbox mode, WhatsApp outbound media from workspace-<agentId> was failing with path-not-allowed because requester agent context wasn’t consistently reaching gateway send paths. not only this path, whatsapp channel is unable to send any media from any path including /tmp.

This PR fixes context propagation in message-tool + sessions_send announce, with tests for both legacy and agent-format session keys.

Would love your confirmation that this matches intended security policy for media root resolution.

[muted badger]

hi guys, anybody come to reveiw this PR made directly from Minimax Team, a user experience improvement:
PR:https://github.com/openclaw/openclaw/pull/33953
Reason:fix(models) Users logging into MiniMax via OAuth will now have the Open Platform's image tools automatically enabled. This functionality has already been implemented for API key users.

[sour sapphire]

Hey everyone, quick update on the LINE requireMention fix —

Tested in a real LINE group and everything's working as expected:

• No mention → bot stays silent, but keeps recording pending history in the background
• @mention → bot responds with the accumulated group context included

So it's not just "going quiet" — it builds up context while silent, then brings it all in when finally mentioned.

Screenshot below, PR: https://github.com/openclaw/openclaw/pull/35847

[pure saffron]

Just to be clear: this is not a test change, but a change to update some types that got broken by a dependency upgrade that broke main. If what you're staying still applies, perhaps consider adding some explicit instructions to CONTRIBUTING.md, because it currently says:

Bugs & small fixes → Open a PR!

... which is what I did

(also, thanks for all your hard work! this is a great platform)

[odd anchor]

tagging @celest merlin @untold pollen since i you guys recently did commits related to gateway.

just lmk or it’ll be helpful to know if the pr is not needed, i am eagerly waiting for next version. thx and sorry

[karmic steeple]

@wise helm, please don't ping Peter for issues, use #1459642797895319552 or #users-helping-users if you need help, use #clawtributors to discuss PRs, or use any of the many other channels in this server as they're intended.

If you have a problem with the Discord specifically, use #report or DM @jolly wolf
-# Your message was reposted above without the ping active for the sake of conversation.

[wise helm]

.

i sound so desperate sorry for this

[brisk elk]

https://github.com/openclaw/openclaw/discussions/38684 Hello community,👋 I have an idea and posted it in the GitHub discussion. Could you please take a look at it? Any feedback would be appreciated.

[wise helm]

https://github.com/openclaw/openclaw/blob/main/VISION.md

[rotund cape]

heyo

[modern kestrel]

Anyone else seeing this? Fast mode on Codex Desktop turns it on / off for all threads in all projects... but if you use App Server via another front-end it can/should be per thread?

https://x.com/huntharo/status/2030338964587180367?s=20

[glossy yew]

@huntharo via Twitter

Harold Hunt (@huntharo)

@OpenAI Codex devs - Running `/fast` in ANY thread in Codex Desktop turns Fast mode on/off in ALL threads? It also seems to cause a problem for older threads that were pinned to GPT-Codex-5.2 or 5.3, they flash the old model name when opened then change to GPT-5.4... I thought these settings were per-thread?

**💬 1 👁️ 19 **

FxTwitter · Mar 7, 2026, 5:45 PM

[empty fjord]

so gng i accidentally left the serv and i lost the role

can i have it back

@clear kraken OwO

[karmic steeple]

Added <@&1458375944111915051> to ! evil german man.

[empty fjord]

ty pookie

[crisp linden]

It seems Deps got updated but pi-ai was missed unfortunately

[jolly wolf]

itll be like *right* when release happens usually

[crisp linden]

Gotcha

[hazy vigil]

hi, not actively contributing to openclaw but i have been hacking away at my own setup. here is something i put together for memory: https://github.com/yoloshii/ClawMem

[vestal pilot]

Hey all I’m an active contributor and wanted to have this PR looked at - it fixes an issue in Discord when sending /new or /reset commands: https://github.com/openclaw/openclaw/pull/38977

[weary notch]

Hey, any particular maintainer who typically enjoys looking at telegram PRs I can tag on a small bug fix or that’s ready for review/merge? (https://github.com/openclaw/openclaw/pull/39172)

[tall prairie]

Take care of those review comments 😉

[weary notch]

Had to do a double take! That last comment I actually addressed it in a792a2a parseInboundMetaBlock now uses continue instead of return null after a parse failure, so it keeps scanning for valid blocks rather than bailing out. The new regression test in strip-inbound-meta.test.ts covers it

Not sure why the clanker made two near identical comments 😅

[modern kestrel]

Join #1474434870259224723 thread for telegram.

I’ve got a PR in to Mac app for Telegram - it hides the autocomplete list for bot commands as soon as you type underscore even though that’s allowed by the spec and works correctly on iOS app. OpenClaw remaps - to _ in bot command names so you can see the problem.

I’m not hopeful though… no commits on main in 8 months.

[weary notch]

perfect thank you

[modern kestrel]

Ohh... my Telegram Exec Approvals buttons is finally clean on checks and merge conflicts:

https://github.com/openclaw/openclaw/pull/37233

[tall prairie]

@modern kestrel can u pull latest main and verify the acp thread stuff you landed is still working?

[modern kestrel]

Yeah. But what do you suspect broke... I've been building something else that I think gives a 10x better experience with Codex.

[tall prairie]

I assume everything is broke

[modern kestrel]

Ok, I'm running upstream/main right now. Give me a few mins.

[tall prairie]

Sweet thanks. I need to get it set up myself so I can start using it.

[tall prairie]

Bruh you are crushing it

[modern kestrel]

https://imgur.com/a/cvtoAxA

It appears that everything is working (except when I mess up sending the messages). I can focus, unfocus, spawn threads. The one error was because I think I like replied and it seems that broke the parsing of the slash command.

[tall prairie]

Nice thank you.

[modern kestrel]

I was inspired to go to the gym today... to confirm I could use Codex the way I wanted while on the treadmill 😂 I have a line item in the implementation plan "gym bro test". Got about 10 minutes in before things fell apart. Tried again while out to dinner... no problems that time.

Here is a little demo of plan responses: https://imgur.com/a/vDE5RzO

[tall prairie]

Sick!!! I love plan mode

[modern kestrel]

/codex_status and /codex_fast toggling: https://imgur.com/a/3nh0v7D

[proud thunder]

Hi maintainers 👋

I'd like to add support for the WeCom channel. My colleague jerryzhou mentioned that your can help with this.

Could you please review and let me know if PR #38976 is still eligible to be merged?

https://github.com/openclaw/openclaw/pull/38976

Thanks in advance!

[modern kestrel]

Check out the join workflow... gives you your most recent Codex Desktop / CLI threads, click to join, gives you last request/response context: https://imgur.com/a/RGgPntn

[zealous plover]

Dang this is awesome

[zealous plover]

Hey @proud thunder, this should probably be a plugin first. https://docs.openclaw.ai/plugins/community

[proud thunder]

@zealous plover It has already been published to npm. Do I still need to submit a PR to this document?

Wecom — Connecting OpenClaw to an wecom bot via WebSocket. npm: @wecom/wecom-openclaw-plugin repo: https://github.com/WecomTeam/wecom-openclaw-plugin install: openclaw plugins install @wecom/wecom-openclaw-plugin

[zealous plover]

How to list plugin: https://docs.openclaw.ai/plugins/community

[zealous plover]

Oh also it looks like the info is on your PR as well in a comment

[pallid summit]

Hi Will I mentioned you in PR but realized I should ask here instead - https://github.com/openclaw/openclaw/pull/32529 can you take a look? 🙏

[zealous plover]

Different Will 😊

[lost kayak]

Hi Maintainers, i have hit the limit of 10 open PR's and unable to contribute further because of new PR limit. Could anyone please review my PR's , here is one of them: https://github.com/openclaw/openclaw/pull/38255

[wise helm]

updated pr time and left a comment for reviewers to have even context,

https://github.com/openclaw/openclaw/pull/38322

It’s a narrow requester/agent-context fix for outbound send paths (message-tool+sessions_send, A2A announce), with partial overlap only on themessage-tool half with #39034.

[proud thunder]

@zealous plover The previous PR (#38976) was closed and cannot be reopened, so I've submitted a new PR. https://github.com/openclaw/openclaw/pull/39511
Thanks.

[keen spruce]

That change to what tools are enabled by default for an agent by [whatever rules] wasn't quite well-thought-out, I'd say. It really leaves people stranded, as there's no indication of why their agents are restricted, especially when it's not a fresh install and stuff just stopped working due to the update. If someone has an idea to handle this better, maybe giving the agent a way to ask the system so it can present proper guidance to the user, that would be great. Maybe a tool like "tool_status" that lists all unavailable tools and the reason why they are not available, and links to the documentation for each reason? (Bonus: Allow the default agent in their 'main' session on the TUI/WebUI channel to check for other agents and sessions so they can implement changes for the user.)

[jolly wolf]

if you're referring to the tools.profile defaulting to messaging, that was a regression that's been un-done in the latest release

[keen spruce]

I don't know what I'm referring to, other than the 937 posts in #1459642797895319552. That's the reason I'm calling this out---even after reading about it so many times, I don't understand it. But if it was just a regression---good. (Hey, I hadn't yet noticed there's a new release.) Still, giving an "effective permissions" report for an agent/session might not be a bad thing. At the moment, there's much guesswork involved in figuring out how all the config values add up, considering there's agent+session+channel+dm/group stacked on each other. And if one of the clawtributors wants to tackle that...

[frail girder]

https://github.com/openclaw/openclaw/pull/33822

Perplexity engineer decides to contribute and "quitely" remove OpenRouter support of Perplexity so that they can benefit more from people using direct Perplexity API instead of going through the universal gateway. Wondering why my agent kept saying 401 unauthorized when doing web search.

[sonic ermine]

sad that it passed the filters and got merged with this regression ... i think you did a good job calling it out , keep up the good work 🦞🦞

[vocal geyser]

I mean you implying intent, when it's possible it was genuine mistake/testing artifact or just conflict with the 2 paths.

[frail girder]

Genuine mistake? i doubt it. He explicitly set base url and model to be deprecated fields. So he must have known what he was doing. Call out @red sparrow to see what he has to say about this

[hazy vigil]

seems pretty obvious that's what he is doing lol

[lean seal]

Just opened a fix for Codex GPT 5.4 context handling in OpenClaw.

Problem:
openai-codex/gpt-5.4 was inheriting a much smaller context window from the gpt-5.3-codex template instead of exposing the native 1,050,000 token window. That meant the Codex path was advertising a far smaller usable context than the model actually supports.

Fix:
The PR patches both the forward compat resolver and the synthetic model catalog so openai-codex/gpt-5.4 now reports the proper native context window, matching the openai/gpt-5.4 path.

Issue:
https://github.com/openclaw/openclaw/issues/39791

PR:
https://github.com/openclaw/openclaw/pull/39797

All CI checks are green and the review thread has been cleaned up.

[vocal geyser]

I am genuinly speaking from experience with the OpenAI TTS. For very long time it had no baseURL as config option and relied on env viarable there were so many strict checks for non baseurl vs baseurl it was nested in directives and commands, I don't know the exact details of the perplexity API and implementation nor the difference with openrouter as provider. In the end openrouter as intermediete still routes to their API. Again my point is not to cast blame and imply intent from code changes, especially with the amount of PRs and confusedopenclaws pushing changes and opening issues and PRs.

[vocal geyser]

Have you considered the Pi-coding-agent 272k context window for GPT-5.4

[lean seal]

Good catch to ask about. The fix does cover the pi-embedded-runner (coding agent) path directly. That is actually where model-forward-compat.ts lives and where the 272k was being inherited from the gpt-5.3-codex template.

The 272k context window is still correct for gpt-5.3-codex itself. The PR only patches gpt-5.4 to expose the native 1,050,000 tokens, and the tests explicitly verify both: gpt-5.3-codex stays at 272k, gpt-5.4 gets 1,050,000.

So yes, the pi coding agent path was the primary target of the fix.

[vocal geyser]

I was hitting my head at this problem earlier today and asked because last release didn't update the pi dependency to the latest we are still 55.3 and gpt-5.4 and some openclaw specific fixes are in the newer releases. And also there are caveats with the higher context. I think above 272k is double the cost and there is serious degradation in performance. Similar fix was suggested by my openclaw but I decided against even in codex it's configuration option gated.

[lean seal]

That is really useful context, thank you. I was not aware of the cost multiplier above 272k or the performance degradation at higher context on the Codex path.

Based on your feedback, I am adjusting the PR right now:

What is changing:

• Default context for openai-codex/gpt-5.4 will stay at 272k (matching the template)
• The native 1,050,000 context becomes opt-in via config (e.g. contextTokens in openclaw.json), consistent with how Codex itself gates extended context
• Model metadata still reports the native capability, but the effective session ceiling defaults to the safer 272k

Why:

• Cost doubles above 272k on the Codex path
• Performance degrades at higher context
• pi dependency is still on 55.3, so some gpt-5.4 behaviour is not fully upstream yet
• Forcing 1.05M as default would be practically worse than the current behaviour

The PR description will be updated with these caveats. The goal was never to force higher context, just to make the behaviour transparent and give users the option when they want it.

PR: https://github.com/openclaw/openclaw/pull/39797

thanks @vocal geyser

[crisp linden]

Well that didn't happen - now the new release is missing those key fixes :/

[red sparrow]

@frail girder Openrouter doesn't support the Perplexity Search API nor the agent API, they only support Sonar, which is outdated, and doesn't provide structured json results directly, but instead AI generated summaries.

If you want to use sonar models, we have an MCP that can do web search through Sonar in a better way than this original implementation: https://github.com/perplexityai/modelcontextprotocol

And yes, ^ supports openrouter

cc @vocal geyser @sonic ermine @hazy vigil

to be clear, if openrouter DID support the Perplexity Search API, there'd be no change made here, there is no benefit to me/perplexity whether or not you use our apis through them or us

The regression in behavior (causing 401s) is a genuine mistake, I should have caught that in my PR

but Sonar should not be a web search option

regardless of where you're calling it from

[frail girder]

Appreciate the clarification @red sparrow . I searched on OpenRouter which perplexity model they support and it seems that they hae full support? https://openrouter.ai/models?q=perplexity. Is it a mistake on their end or is the perplexity search api not supported in those models? Curious to know. Thanks

[red sparrow]

Every model you list on this page is a Sonar model, the Search API isn't a LLM, it's direct json results

https://docs.perplexity.ai/docs/search/quickstart

They also list models like Perplexity: Llama 3.1 Sonar 70B Online which do not exist in our API anymore and were deprecated ages ago, we're coordinating with openrouter to resolve that

I'm happy to make another PR to add a deprecation warning in the OpenClaw UI that tells them to use perplexity directly or the above MCP if that'd help, I'll reiterate: there is no way I or perplexity benefits from you going to us directly as opposed to using openrouter (since you're getting 401s now, we're missing out on your usage through OR which is technically worse for us )

[red sparrow]

example usage attached, this is MUCH faster and skips the step of having one LLM talk to another to get the same information

and since web_search defaulted to sonar-pro, you were limited in OpenClaw regardless

via MCP, you can call any model, including deep research

hope that clears everything up, I'm more than happy to answer questions and genuinely love openclaw haha

[frail girder]

Hmm, then how did the previous option with openrouter work in the first place? Cause that must have worked right? The previous one call the Sonar model to make the search right? Perplexity Search API just makes it easier by actually calling the api and return nicely json output right? Reason I'm asking is because I want to use only openrouter, managing too many api keys is a pain

[red sparrow]

yes exactly, openrouter called sonar-pro which led to a multi turn web search (that gave you basically the same results) but than a model had to ground that information into a summary with citations (much like how grok, kimi etc work as a option)

this wastes tokens summarizing twice, with sonar pro and with your claw. If you want to keep using sonar pro, use the MCP

also, happy to send some api credits your way (if you're comfortable with another key in your config lol) as a gesture of goodwill, I never intended to worsen the openclaw experience

[frail girder]

No worries mate. Seems like perplexity up their game but OpenClaw doc hasn’t caught up yet. Thanks for the PR and for the updated info here. Cheers

[vocal geyser]

more likely openrouter issue tbh

[grizzled rivet]

Hey clawtributors, looking for eyes on a plugin hook PR:

https://github.com/openclaw/openclaw/pull/20067

Adds before_agent_reply — a plugin hook that fires after slash commands but before the LLM runs, letting plugins return a synthetic reply to short-circuit agent processing. Enables forms, approval gates, wizards, and interactive tutorials as plugins without touching core.

There are 60+ open issues/PRs requesting some form of message interception, and none of the existing hooks (including the newest ones from this week) fire at the right pipeline position to do this. The PR body has the full breakdown.

[untold pollen]

Ah, I wish I saw this earlier. I just landed a PR that restores OR compatibility on #39937

[red sparrow]

maybe be good to revert? apologies

[wind crag]

@jolly wolf Claiming my contributor role - my PR #39281 (fix: reject pid <= 0 in parseLaunchctlPrint) was landed on main by steipete.

https://github.com/openclaw/openclaw/pull/39281

The fix stops launchd runtime parsing from recording sentinel/non-running values as real PIDs, with regression tests for pid = 0, positive, and negative cases.

GitHub: @wind crag

[tall prairie]

Please don't ping Shadow for that. Mods will grant you the role.

[untold pollen]

I don’t think an immediate revert is the right move. #39937 is just restoring broken existing OR/Sonar configs and fixing the 401 regression (and not claiming that Sonar == the Search API)

Better follow-up, IMO: keep legacy compatibility for existing users, add a clear warning/deprecation for the OR/Sonar path, and steer new setups to direct Perplexity Search API or the MCP

[red sparrow]

looked at the changes, LGTM, agree this is the right approach! Thank you.

[wind crag]

sorry. it was Shadow who told me to claim it on X

[tall prairie]

we ship fast and break things, no harm no foul. Thanks for being active @red sparrow

[grizzled rivet]

not sure what that approval process looks like but putting my name out there for contributor role eligibility (#1917 #18218 @grizzled rivet on GitHub)

edit: yay! thank ya kindly

[vocal geyser]

is there a proper way to communicate and discuss things with a maintainer and their specific responsibility I just don't feel like tagging them directly or DMing and submiting just random PRs that half ass fix something doesn't feel right to me and I'm already 20+ hrs deep in their specific field/implementation

[untold pollen]

Most maintenance are listed here with their area of responsibility: https://github.com/openclaw/openclaw/blob/main/CONTRIBUTING.md

There's also channel specific topics, e.g. this one for Telegram: https://discord.com/channels/1456350064065904867/1474436078373966089

[wise helm]

I found what looks like an OpenClaw sandbox/browser design bug. A sandboxed agent can use the browser, but browser screenshots are saved to host media storage, and the sandboxed agent
then can’t send that file back via message because the file path crosses sandbox/host boundaries. Is this a known issue, and would you want a PR for it? If yes, what’s the intended fix:
shared media/artifact handoff, looser message path handling for host media, or a different browser-output design?

[vocal geyser]

I know who the maintainer is there is no specific channel for the acp/acpx work he did, i also read the whole implementation plan in docs/experiments/plans etc I can easily create multiple issues but this will only trigger 100 automated PRs without actually understanding the issues so maybe #architecture is the best channel for such discussion if onur even uses the discord

[wise helm]

i am not talking about one specific person but yeah this definitely feels like talking to a wall

maybe another claw can help better lmao

[vocal geyser]

I mean we are 100k+ members and there are only so many maintainers that's the main reason i don't want to ping/DM bug them for no reason cuz i can imagine 500 people doing it

[wise helm]

500 is less imo. also i think they of-course have a channel and of course i think not all maintainers are on discord. the problem is design direction and alignment. for me it's been 30 days. one version introduces a bug; the other solves it but another bug is introduced.

[vocal geyser]

that's expected and that's why i'm looking for proper way to communicate what can be useful and correct instead of just fast PR spam like most people are doing

[wise helm]

us

[final ocean]

im looking into the full hooks surface area over the next few days

[grizzled rivet]

heck yeah. let me know if there's anything i can do to help or other PRs you request. happy to throw a disgusting amount of tokens towards PRs

[sharp ivy]

Hi @grizzled rivet Looking at your PR! Thanks for the link to mine -- very helpful!

@grizzled rivet Nice and comprehensive overview of the existing work -- it looks like you're adding yet another critical feature not found anywhere else -- would love to help out in any way I can! Learning how to contribute here, so feel free to tell me where I can be of assistance 🙂

[silent bolt]

Hi there , i had submit PR about a2a im solution https://github.com/openclaw/openclaw/pull/36250 as channel plugin . Should i talking to anyone for merge it?
Related info :
https://github.com/openclaw/openclaw/discussions/857
https://github.com/Prismer-AI/Prismer/tree/main/sdk
https://github.com/VoltAgent/awesome-openclaw-skills/pull/203
Welcome to chat to about any question.

[keen spruce]

Just found this one in my notifications: https://github.com/openclaw/openclaw/pull/19736#issuecomment-4017010228 Seems the "select which PR for the same problem to merge" process selected one that doesn't work. I could send the user who actually tested it to open an issue, but yeah, let's not invite even more PRs...
Edit: Matrix plugin

[spiral swallow]

Hey! Just joined — I'm an OpenClaw user and I've been digging into the iOS app from source. Already went through the apps/ios README, the xcodegen flow, and the current node command set (camera, canvas, screen record, location, Watch App).

I'm not a Swift specialist, but I love building things and I tend to dig in hard until I figure it out. I use OpenClaw on iOS daily and I care a lot about the iOS node experience — so that's where I want to contribute. Happy to build and test on my own device, report issues in detail, or pick up tasks on the iOS side. Is there a good place to start?

[jolly wolf]

#1458485303038836851

[pallid summit]

Hey all, looking through the contributing doc, it doesnt mention what to do for PR not getting reviewer after some time. Should I ping reviewer directly, share PR here, or what? I notice people doing that already and old (meaningful) PRs are still not getting maintainer attention. Wanna be sure to not spam maintainer unnecessarily. Thanks!

[brisk frost]

I am trying to solve a problem and CI failures are pre-existing and unrelated to my PR:

detect-secrets: Baseline file is out of date — flagged files (matrix.test.ts, AppStateRemoteConfigTests.swift, etc.) are not part of this PR.
loader.test.ts: "supports legacy plugins importing monolithic plugin-sdk root" timed out at 120s — flaky test on Windows runner, not related to model resolution changes.
Can anyone guide me?

This is my PR
https://github.com/openclaw/openclaw/pull/40211
Just check it out

[inland field]

im using it really cus my entire family uses my openclaw and to much of a pain to setup 5 different instances

cus i want all the memory channels and cron jobs to be seperate

while using the same apis and everything else

[crisp linden]

@jolly wolf please check

[smoky lintel]

@dusky blaze claiming my contributor role for PR “feat(gateway): add trusted-proxy auth mode”, https://github.com/openclaw/openclaw/pull/15940

It got merged back in mid-February. Coming a little late to claim my role. 😅

[karmic steeple]

Added <@&1458375944111915051> to nickytonline.

[dusky blaze]

Congrats!

[smoky lintel]

Thanks!

[smoky lintel]

ty!

[past tinsel]

Hey, I opened a small fix PR for the BlueBubbles packaged-install issue plus the remaining bundled-plugin src/* import cleanup in Mattermost, Nextcloud Talk, and Twitch. CI is green if anyone wants to take a quick look: https://github.com/openclaw/openclaw/pull/39638

[tulip tusk]

I created a couple of fixes for WhatApp pairing which is currently broken. More details here in the channels section:
#1474436109424525637 message
Let me know if I need to do anything to get these merged

[red sparrow]

@untold pollen could you PTAL at https://github.com/openclaw/openclaw/pull/40259 ? not significant for what we discussed earlier but a nice to have

[tulip tusk]

Fix to include web search providers in the SecretRef coverage
https://github.com/openclaw/openclaw/pull/40511

[untold pollen]

Merged! Thank you

[tall prairie]

is this the one that works?

[red sparrow]

@jolly wolf clawdhub is down, I think it's an env vars thing

[CONVEX Q(appMeta:getDeploymentInfo)] [Request ID: 0911c0a29382629b] Server Error Called by client

https://github.com/openclaw/clawhub/actions/runs/22830800173/workflow

quick PR to fix: https://github.com/openclaw/clawhub/pull/639

I saw in maintainers page you're DRI for this hence the ping

[untold pollen]

Just pushed a patch to make sure that single error doesn't bring the site down. Looking into the root cause

[red sparrow]

my guess its this from peter's CI hardening, issues have existed for 5hrs https://github.com/openclaw/clawhub/issues/634

clawhub.ai is now just a (pretty) pink for me, and deployments are still failing

im p sure its just the secrets thing my pr covers

[ocean hill]

hey folks! i had a PR merged, who do i request a contributor badge from?

https://github.com/openclaw/openclaw/pull/31044 my gh is barronlroth

[red sparrow]

@exotic bloom mind checking https://github.com/openclaw/openclaw/pull/40607 ? small but should fix the nit lol

[feral prism]

@untold pollen Let me know if I need to do anything on my end https://github.com/openclaw/openclaw/pull/39388

[untold pollen]

I think you replayed commits here. This is a 20K line diff.

[feral prism]

on it

[zinc gorge]

Hello

[feral prism]

@untold pollen Cleaned https://github.com/openclaw/openclaw/pull/40636

I realize it's a ton of code - it's all sanitization specced.

[untold pollen]

Sorry, can't review a 18K line diff right now.

[feral prism]

all good understood =]
just making sure I had what was needed

[languid folio]

did a bunch of triage and testing for open issues that I couldn’t replicate the bug. already tagged peter in the comments for the all the issues that can be closed because they are already fixed in the latest release

should I copy the issues here so other maintainers can close them?

[untold pollen]

Sure, I can take a look and close them

[languid folio]

https://github.com/openclaw/openclaw/issues/40512
https://github.com/openclaw/openclaw/issues/40507
https://github.com/openclaw/openclaw/issues/40559
https://github.com/openclaw/openclaw/issues/40545
https://github.com/openclaw/openclaw/issues/40550
https://github.com/openclaw/openclaw/issues/40630
https://github.com/openclaw/openclaw/issues/40551
https://github.com/openclaw/openclaw/issues/40563

@untold pollen 40551 and 40563 I can’t replicate to confirm locally since i don’t have a windows i can test with but the code fixes seems good

let me know if i missed anything

[dark needle]

HI team, may I have a review on https://github.com/openclaw/openclaw/pull/38695
The PR integrates SGLang-Diffusion to openclaw, building the image-gen tool

[wise token]

Hello there. PR ready for review - adds WhatsApp SystemPrompt hierarchy (account+group level) and fixes account-level group policy. CI green and Codex comments addressed. Link: https://github.com/openclaw/openclaw/pull/40250 Thanks!

[tiny smelt]

Hi all. I have 3 PRs ready to go.
https://github.com/openclaw/openclaw/pull/39308
https://github.com/openclaw/openclaw/pull/39410
https://github.com/openclaw/openclaw/pull/39290 [<< this one may already be fixed]

[zealous plover]

It looks like Peter already pushed a fix for this

[red sparrow]

yes, after it was flagged

[oak fiber]

hello，anyone could grant me clawtributor role? theses are my merged PR: https://github.com/openclaw/openclaw/pull/38533 https://github.com/openclaw/openclaw/pull/13818 , could you? @karmic steeple

[gleaming bay]

Hey — quick question about PR attribution when closed PRs get resubmitted.

My PR #40615 fixed the EBUSY update issue (#40540) and was closed by barnacle bot (too-many-prs). PR #40724 extracts the same fix and references my PR in the description, but no Co-Authored-By in the commit. I asked in the PR comments — no response yet.

Similarly, my PR #36755 fixed groupAllowFrom negative IDs (#36753) on Mar 5. Closed by barnacle. PR #37134 was merged Mar 8 with the same fix — no reference to my work.

Is there a recommended process for attribution when closed PRs are resubmitted by others? Happy to keep contributing, just want to make sure credit lands in git history where it belongs.

[dry loom]

Hi everyone, KIMI model tool calling is broken since version 3.7 , many people are influenced. I opened a PR #40008 to fix it, who can please to push the process to check and merge?
Thanks.
I think it's QUITE URGENT : )

[white galleon]

if your PR got merged and not credit please reach out to whoever merged the PR. If someone else's PR fixing the same problem got merged instead of yours it's probably because their code provided a better baseline for however the maintainer judged the fix should be implemented; but if someone is literally copying your PRs then please let us know here and we will ban whoever is copying other people's PRs

[pallid summit]

Hey - I've submitted https://github.com/openclaw/openclaw/pull/32529 for a week and seems there's no review on it yet. Am I missing anything here? Would appreciate a quick review.

[plain olive]

Hi @untold pollen , I've submitted a PR: https://github.com/openclaw/openclaw/pull/35571 to add a new search provider querit.ai. As requested in the CONTRIBUTING.md, I've also started a discussion thread here: https://github.com/openclaw/openclaw/discussions/36053 .

Could you please take a look when you have a moment? Any feedback would be greatly appreciated. Thanks!

[languid folio]

guys we need to write better issues too… a great amount of the open issues I’ve been testing have false root cause identified

and people who use coding agent to fix them all the way without reviewing at all do monkey patches for it just to leave another mess for other people to clean up

[karmic steeple]

Added <@&1458375944111915051> to barron.

Added <@&1458375944111915051> to xialongLee.

[compact stag]

Hi, is there anyone could review and approve this pr? https://github.com/openclaw/openclaw/pull/37422

[oak fiber]

thx

[untold pollen]

Hi, if any of you are still facing Telegram duplicate message issues, please feel free to report them over in: https://discord.com/channels/1456350064065904867/1474436078373966089

[tall prairie]

You’re manually testing issues to repo and the information in the issues is insufficient? Can you send a small sample of these issues?

[dry loom]

Hi everyone, there are already many people reporting this issue on social media. Is there anyone who can help merge and fix it ASAP? 😕

[wicked viper]

Bug Fix PR: ACP session spawning broken on v2026.3.9 🐛

Ran into a bug where ACP session spawning via sessions_spawn (with runtime: "acp") fails with:

spawnedBy is only supported for subagent:* sessions

Root cause: sessions-patch.ts validation for spawnedBy/spawnDepth only checks isSubagentSessionKey(), but ACP sessions use agent:*:acp:* key patterns which don't match.

Fix: Added the existing isAcpSessionKey() utility check alongside the subagent check at both validation points. Minimal change, all 20 existing tests pass.

• Issue: https://github.com/openclaw/openclaw/issues/40984
• PR: https://github.com/openclaw/openclaw/pull/40981

[vocal geyser]

btw there are at least 3-4 other PRs for this already first one i could find is https://github.com/openclaw/openclaw/pull/40316

[keen spruce]

Considering the PR is still mergeable, I would assume so.

But this really makes me question what the original change wanted to accomplish. kimi-coding had been working fine for ages (I can personally confirm that)---why change its parameters? That's inviting breakage. We really need to get from "fix" PRs getting merged because they state they fix something, and finding a way to confirm they actually do and don't just smash the china near a suspected problem. AI code submissions are great for new features and internal bugs, but when it comes to bugs that are about interacting with stuff that is not in the codebase, AI's have to rely on guessing how it should behave. Whether it's a "security fix" that misidentifies the operating system as an attacker, a "fix" that guesses how Matrix transmits chatroom data, or one that hallucinates what parameters to send to a provider. PRs that touch interactions with the outside world need to be verified and tested by a human.

[tall prairie]

e2e llm tests are possible just expensive.

i'm considering introducing a test suite that is private because llms change tests to make their changes work all the time it kind of defeats the point

sort of like the sqlite model, code is open source but the test suite is private

https://tenor.com/view/every-time-60percent-60percent-of-the-time-it-works-every-time-anchorman-gif-4630682

for as much as things change, it mostly works

[keen spruce]

then at least ask people who use that model/provider to confirm that it is currently broken before fixing it. I could have told you in a second that tool calling with kimi-coding worked. I can't find on github what prompted that change, it looks like a random commit with no trail to any reason, so I don't know why it was changed. But if someone reported broken tool-calling, then that would have been the right moment to investigate why it wasn't working for them when people have been using that model for weeks.

Yes, sometimes changes break things---I have broken stuff and even released versions with such breakages often enough myself. But there are fields where you double-check if there's (a) actually something to fix and not just a "problem between keyboard and chair" issue and (b) that the fix works

[keen spruce]

I'm not calling for the "big and impossible solution to life, the universe and everything". I'm asking to step back and think for a minute when looking at proposed changes. Can this break anything? If so, who can assess that before we merge/commit? Is the bug this should fix even real? Do we have someone to confirm the bug and the fix? --- Simple mental checklist that takes 10 seconds.

[tall prairie]

i can't chase down every bug and conduct interviews with the entire userbase. i have to assume if someone comes here, posts, they work through the issue, claim verification, there are several related issues then it's tested.

pull main, test it and if there is a problem open a pr and post it here.

[untold pollen]

Honestly at this scale there's just so many things breaking it's hard to isolate / reproduce issues to a single problem

[untold pollen]

I manually tested several configs, environments (groups, DMs, forums) on 3 different platforms (WSL2, macOS, Docker) with 2 different LLM providers (Claude, Codex) and exhuastively confirmed that the duplicate message bug is fixed in Telegram

And we still have users facing it

[tall prairie]

too many permutations...

[keen spruce]

Sorry, didn't want to make a big discussion out of this. I'm just personally annoyed because I want to finally update my installation, but there has been a regression that would affect me in every single version for quite a while. And, honestly, the project isn't at a stage where we can afford people ignoring updates because they're afraid of breakage. This is not notepad++ where the safest thing actually was ignoring updates for years ;)=

[untold pollen]

That's completely understandable 😅

We try to test every PR before merging but things still slip in

[keen spruce]

yes, trying to fix a confirmed bug and having to try and try again because it pops up in different setups...that's something else. I had a couple of these bugs where I eventually had to write code that simpy recovered the game state when they happened as I could never find them... (I maintained Ender-IO from Minecraft 1.8 to 1.12. Not to brag, just as a reference)

[vocal geyser]

Yesterday's commits broke Ran into a bug where ACP session spawning via sessions_spawn (with runtime: "acp") fails with:
spawnedBy is only supported for subagent:* sessions
And even on the PR codex-review pointed that out but it still got merged

https://github.com/openclaw/openclaw/pull/40137#pullrequestreview-3911844669
here is the codex-review on the PR

[grizzled rivet]

I fixed another small build bug

The macOS build has a Swift file (HostEnvSecurityPolicy.generated.swift) that's auto-generated by a script. Right now, swiftformat and swiftlint both process it, which means every time the file gets regenerated, the formatters rewrite it, creating noisy diffs.

This PR adds it to the exclusion lists in .swiftformat and .swiftlint.yml - same pattern already used for GatewayModels.swift (another generated file).

https://github.com/openclaw/openclaw/pull/39969

(greptile 5/5)

[final ocean]

Merged

No changelog since its a ci change

[grizzled rivet]

sweet

btw - is this annoying/helpful to you? plugging PRs here?

ideally i find most impactful things to work on and don't needlessly add noise, not super clear to me what the community expectations are

[tall prairie]

yikes

[final ocean]

You don’t need to ask this question if you read the repo

[arctic siren]

or use the Chat

[modern kestrel]

Do you have example commands? What channel is this in? Is this in Telegram using like /acp spawn (if so... I worked on that so I can test it)

[vocal geyser]

I was using discord I mean there are multiple issues and PRs for this
I triggered this when testing a PR to trigger thinking/model but should be universal for anything trying to spawn acp

[neat elk]

fixed with this PR https://github.com/openclaw/openclaw/pull/40995

[pale crane]

Fixed a couple small ACP issues based on SuperagenticAI article I saw on X. cc/ @versed kindle

https://github.com/openclaw/openclaw/pull/41185

https://github.com/openclaw/openclaw/pull/41187

[versed kindle]

you can tell the agent to use acpx directly, acpx was originally built for this purpose

but there should be first class support for this already I think, what happens when you ask it to create acp subagent without thread?

[versed kindle]

can you share the article btw, I can't find it

[pale crane]

https://super-agentic.ai/resources/super-posts/openclaw-acp-what-coding-agent-users-need-to-know-about-protocol-gaps

[magic vine]

Maybe I’m not using the right words, but it keeps routing it to threads/output to me and says it can’t see it

[pale crane]

ps. I used ACP to create these PRs 😉

[versed kindle]

nice, telegram?

[pale crane]

yup telegram on 2026.03.07 and it ran parallel sessions with Codex

[versed kindle]

where did this get featured btw, I saw Peter share it earlier as well. only I haven't seen it apparently lol

[pale crane]

it showed up on my timeline based on Peter's comment. this is the tweet https://x.com/Shashikant86/status/2030931454763430221

[versed kindle]

@pale crane I'm landing both PRs now, ty!!

[pale crane]

amazing. happy to help on other bits, just wanted to do a quick pass. I'm a user of ACP so feel free to ping whenever you need help or another tester

[versed kindle]

did you try to tell it to use acpx directly with its bash tool, and bypass openclaw acp setup?

acpx has a bundled skill, you can also ask it to install the skill from acpx --skill

[magic vine]

Ok I’m having it check

[red sparrow]

small pr https://github.com/openclaw/openclaw/pull/40607

[languid folio]

manual tests for some, some even llm e2e tests for models that i have configured. most i can only make sure test suites pass. for example the proposed fix for https://github.com/openclaw/openclaw/issues/41246 would not work

[keen spruce]

Can you please link your GitHub account in your Discord account? It's just a couple of clicks for you and helps us believe you that both are by the same person. Thanks.
That's "Click on your user name -> Edit Profile -> Connections", btw.

[left mango]

I have added the connection now, thanks

[karmic steeple]

Added <@&1458375944111915051> to normie.

[keen spruce]

Then here you go.

[left mango]

Thank you! Just out of curiosity - if my github account would have been connected to my discord account, would the bot have given me the role automatically?

[keen spruce]

No, it wouldn't. We don't have that enabled because there are some ways to misuse it that I won't go into here.

[magic vine]

Bash worked, thanks! I had it do a write up of what was tried before and the issues in case this is of interest https://gist.github.com/davidguttman/963e733b0143dcdc4c6c31220f2d71df

[versed kindle]

I think there should be a native way to spawn an acp session with thread: false, which is equivalent to just calling acpx over bash

If this doesn't work, would you like to send a PR to fix it? Not sure how complicated this is. If sessions_spawn needs a large refactor, the interface should be scrutinized

[wise meteor]

@keen spruce can you add me to the clawtributors role group?
https://github.com/openclaw/openclaw/pull/32108

[karmic steeple]

Added <@&1458375944111915051> to miotch.

[wise meteor]

awesome. thx

[smoky lintel]

Tiny doc update related to trusted proxy auth https://github.com/openclaw/openclaw/pull/16251 . TLDR; removed one line in the docs.

[pallid summit]

Followed the process to start the discussion at https://github.com/openclaw/openclaw/discussions/41360 with PR linked. PTAL thanks!

[vocal geyser]

@versed kindle i submited a PR that solves the model/thinking at least for Codex on acpx. https://github.com/openclaw/openclaw/pull/40370 would be glad if you can take a look and give me some feedback,

[pale crane]

opened a third PR from that SuperAgentic article. this one scopes cancellation and event routing by runId so multiple ACP sessions on the same gateway key don't cross-wire.

Note there is a CI failure on oxfmt issue in CONTRIBUTING.md. Unrelated to this PR but happy to fix too if it makes it easier to merge/review.

https://github.com/openclaw/openclaw/pull/41331

[weary notch]

submitted a PR for #33554 to fix Discord extension outbound sends ignoring threadId in extensions/discord/src/channel.ts, no revisions needed according to the clankers. https://github.com/openclaw/openclaw/pull/41388

[marsh seal]

I know it's hard to get to features these days, but I have had https://github.com/openclaw/openclaw/pull/36901 up for about a week, any idea how long it will take for a review?

[bronze pike]

Hi there peeps. I've been wondering about something -- I think it'd be nice if github orgs or repositories could have guidelines for how autonomous agent should interact with them (really not specific to github, perhaps any website).

For example, these might include: When opening a PR, you MUST identify that it was built with openclaw and provide a summary of the instructions from the user. After opening the PR, you MUST add a comment@triagebot ai-authored to flag it as AI-authored. or similar things.

Context is that I'm coming from rust-lang/rust, where many reviewers are struggling from a lot of auto-generated AI PRs, some written with agents, some generated entirely autonomously with openclaw or other similar tools. I noticed that y'all have a limit of 10 open PRs for similar reasons.

I am curious if openclaw has something like this, kind of like AGENTS.md but for websites and things you are interacting with, and/or has tinkered with policies like this.

[keen spruce]

In theory, agents should read the AGENTS.md file (e.g. https://github.com/openclaw/openclaw/blob/main/AGENTS.md). In practice...that doesn't work so well. The issue is mainly that agents are universalists, so it's up to the user to give them specific guidelines for specific tasks that the LLM models haven't been trained on. Pure coding environments are better at this, e.g. Claude Code will force the LLM to read CLAUDE.md (which is a symlink to AGENTS.md in our repo).

[keen spruce]

Can we get a maintainer decision on whether we want search providers as plugins or not? That poor guy has been updating this PR for over a month now, and even someone else now made a second one because it once got closed because when it was misclassified as a plugin instead of a plugin API. If it's not wanted, just release them from their pain, please.

Any plugin can hijack search: https://github.com/openclaw/openclaw/pull/41288
Search Provider as plugin type: https://github.com/openclaw/openclaw/pull/11444 (better in my opinion)

[wind crag]

Is there a way to get flagged to allow more than 10 open PRs at a time? Had a few of mine get closed that were good ones. Just had a user ask if I can re-open this and the answer is no bc I have 10 in queue? I'm a Clawtributor on here. https://github.com/openclaw/openclaw/pull/40595#issuecomment-4027439719 is the one that's not even in the queue bc I have 10 others that the user saked if I could bring back

[left mango]

On a side note, I checked June Oven and it seems insanely good!

Have you posted anywhere how you/ June Oven are using OpenClaw? I would love to check it out.

[wind crag]

@left mango Haha, no, we sold the company and we took a very Apple-like approach, so we had a very closed ecosystem. But it would be fun in the land of OpenClaw to control my June Oven with it. Alexa was the best we got.

[left mango]

I would definitely let openclaw control my oven when my wife and kids aren't at home 😂

[glossy python]

The Brave LLM change in 2026.3.8 is broken unfortunately.. I submitted a bug report (https://github.com/openclaw/openclaw/issues/41370) and there’s a PR created (https://github.com/openclaw/openclaw/pull/41387). could someone review it please?

[dark needle]

HI team, may I have a review on https://github.com/openclaw/openclaw/pull/38695
The PR integrates SGLang-Diffusion to openclaw, adding its first image-gen tool

[keen spruce]

This would mean that the original PR never worked and hadn't even been tested once by the submitter? Are you sure this is an "always" problem and not an alternative constellation?

Oh, forget it. The original PR even says so: "What I did not verify: Live API call to Brave LLM Context endpoint"

[left mango]

Is this a fair comment?
https://github.com/openclaw/openclaw/issues/41674#issuecomment-4028338048

If yes, the bug report can be closed

[soft flicker]

@keen spruce what are qualifications to get added to contributors? I've had a couple of feature PRs, but nothing picked up yet.

[soft flicker]

https://github.com/openclaw/openclaw/pull/39126

https://github.com/openclaw/openclaw/pull/39244

@jolly wolf

[jolly wolf]

You need a merged PR

[soft flicker]

Oh. I do not have any of those yet, lol.

I feel like my first one (trust windows) is a strong feature, but I'm not sure how to ensure it gets visibility on the 5k list.

Second one is solid as well (actually useable implementation of Safebin), but I'm less confident in it.

Both are based around the idea of making security easier for folks to use so they don't just turn stuff off or overpermission their bots.

[jolly wolf]

gonna be completely honest, that first one is shit

we have a command registry for a reason, don't bodge some ai slop into just a single channel

[soft flicker]

Honestly, thank you for review.

I'm absolutely vibe coding it, so I'm sure the code quality is junk.

But I'm curious your critique of the user model. Is it that trust windows are the wrong approach, or is my view of it just sloppy?

[jolly wolf]

the idea is good, your implementation is awful

what ai agent are you using

[soft flicker]

Well, that's what you get when you have a lifetime Product Manager who thinks he can vibe code 😉

Just vibe coding with Openclaw, using Opus 4.6 and various gpt 5.3 codex agent subs for review.

[jolly wolf]

close the pr, start from scratch, and use codex entirely for coding it. slopus is too narrow minded

[soft flicker]

That works. Would give me a chance to add a proper TUI implementation as well.

Already solved a bunch of the tough design challenges to prevent the bot from being able to extend or defeat the escalation while escalated.

Got a spec I'm working from as well, so can work with codex to implement what's already speccd.

[shell dawn]

sorry a bit of an aside, 5.3 codex over 5.4?

[jolly wolf]

personally yes

5.3 codex fights against bad decisions more imo

[shell dawn]

i only did a little bit of 5.3 codex before i got 5.4, i'll have to try it out more. getting a little tired of the 10 volume compedium for my idea on making 3 changes

[compact stag]

@karmic steeple Hi, is there anyone could review and approve this feature pr? https://github.com/openclaw/openclaw/pull/37422

Where can I get the SOP of merging a pr?

[hoary pendant]

the more fixes based on the greptile&codex review comments I add ... the worse the greptile score becomes 🙂

[untold pollen]

Codex review comments are really good though 😄

[valid jay]

hi everyone

is this the right place to share skills?

[untold pollen]

#skills

[valid jay]

can i share something i just built?

[pulsar arch]

Could I please have feedback/review/merge of https://github.com/openclaw/openclaw/pull/38930 - "feat(retry): expand global retry codes and migrate Venice discovery". Thanks 🙏

[coarse stag]

Hey mainteners, I opened a PR fixing a deterministic race condition in streaming voice calls (Twilio + media streams). Basically the voice plugin isn't really usable with streaming enabled without this fix. (!)
Small, isolated fix (~25 LOC) in the voice-call extension only.
PR: https://github.com/openclaw/openclaw/pull/40170

[viral rock]

There was a bug in the Discord integration where SecretRefs would get lost, in my case causing errors when an agent tried to create a Discord thread (channels.discord.token was not found). I fixed it in https://github.com/openclaw/openclaw/pull/41142

Very strange to be in this new era were PRs are being spammed as self-promotion tactics, mixed feeling about the opensource community right now. I hope this doesn't break the spirit of real contributors but you all seem to be managing well 🙂

Anyway: I would be surprised if a PR like mine will actually be merged. How do you all handle keeping up to date and having multiple PRs you need on your local system on top?

I now use a local integration branch and a skill to manage updating from upstream, rebasing my PR branches and cherry-picking them in my local integration branch. Is there a better way to handle it?

[viscid lotus]

Which branch is used for releasing a new version?

[viral rock]

main I guess, if the git branch is what you refer to?

[novel monolith]

Good morning, I found a bug with ACP that cost me $20 to fix. The problem is that I have codex with my sub but when openclaw spawns codex via acpx it is not stripping api keys, thus codex finds the OPENAI_API_KEY and overrides the auth. It depleted my credit in 2 prompts.
Can anyone take a look at this? I think this can be impacting people and it's hard to sort out: https://github.com/openclaw/openclaw/pull/41615

[left mango]

Hi, I just created this PR to solve an issue reported by someone for Slack (Slack route binding peer ids).

https://github.com/openclaw/openclaw/pull/42162

[lost kayak]

Hello Maintainers, i am unable to create new PRs as i have hit a limit of 10 Open PRs again. Could someone please look into my PRs? Here is the most recent one: https://github.com/openclaw/openclaw/pull/41161
Thank you

[fossil adder]

May joy and abundance find you as well!

[true obsidian]

If you are building a skill this may be an issue for you during publish. https://github.com/openclaw/clawhub/pull/685

[pallid summit]

Followed the process to start the discussion at https://github.com/openclaw/openclaw/discussions/41360 with PR linked. PTAL thanks!

[tardy fossil]

Hey 👋 Quick question for the maintainers — I have a ClawHub skill (domain-trust-check) that lets agents check URLs for phishing or malware before visiting them. It’s basically just curl + an API key, similar to how the Notion or Trello skills work.

I was thinking about proposing it as a bundled skill, since URL safety is a security area that everyone seems to be ignoring, but it’s a huge attack surface for agents. I believe this will be extremely important as attack against agents are also scaling. What do people think?

GitHub Discussion: https://github.com/openclaw/openclaw/discussions/42420

[wide panther]

Hey everyone, I finally cracked the NotebookLM 'Authentication expired' login loop bug that's been breaking our agents for days!

The issue was Playwright's SameSite: Lax security policy dropping Google's rotating session cookies. I built a custom, drop-in 'Self-Healing' skill that completely bypasses this. If the agent gets an auth error, it now automatically launches a visible browser for 5 seconds to bypass bot detection, steals fresh cookies, injects them directly into the Windows environment variables, and retries the task silently without you ever having to type notebooklm login again.
You can grab the full skill repository and drop it straight into your workspace/skills folder here: https://github.com/antaripnandi/openclaw-notebooklm-skill

[languid folio]

hey, noticed #36838 (gemini-3.1-flash-lite-preview support) is still open — the google provider got the fix but google-vertex was missed. someone left a comment on one of my earlier PRs pointing out google-vertex/gemini-3.1-flash-lite-preview wasn't working (#36918 comment), so i looked into it.

threw up a quick fix in #42435 https://github.com/openclaw/openclaw/pull/42435 — just extends the existing normalizeGoogleModelId() to also cover google-vertex in the two places it was missing (model ref parsing + provider config normalization). 4 files, 43 lines, all tests passing.

[pallid summit]

@untold pollen mind taking a look at my PR? Appreciated https://github.com/openclaw/openclaw/discussions/41360

[slim vessel]

Put it in #showcase-old

[hoary pendant]

I did a thing ... I know you don't want things ... but I did it anyway.

Support for media plugins = https://github.com/openclaw/openclaw/pull/41496

I am also writing such a plugin, and testing it, so I'll understand if none of you fine folk want to look at my large-ish PR before I can actually prove that it works for me.

[left mango]

The original issue seems like a nice feature request and I have added a comment there. I did not go ahead and create a PR because this requires a buy-in from maintainers.
https://github.com/openclaw/openclaw/issues/41657

[jolly wolf]

Go ahead, try to make something channel agnostic, though, not like that “kind” variable you have, because we have to support a theoretically infinite number of channels because plugins can add channels

[versed kindle]

can you dm me this? lets discuss there

[vocal geyser]

@versed kindle sent friend request

[versed kindle]

ty, will look into it tomorrow, about to sleep

Hi everyone

Request for beta-testers

For a new stealth openclaw based project

I am looking for companies that have deployed openclaw internally, but are bottlenecked by having just one agent for everyone on Slack, Discord, Teams, Mattermost, etc.

If you qualify, please reply to this or DM me

Note: this is not affiliated with openclaw, but this is something I've been building since 2 months at textcortex

[hoary pendant]

You know that you can map agents to topics (in telegram at least) ... right?

[versed kindle]

I know, I've built the acp bindings for that

[dawn light]

Hi @jolly wolf, why PRs are not getting reviewed or merged. I am seeing a pattern only maintainers PRs are getting merged.

[jolly wolf]

Yes? That’s how pretty much all of open source works, maintainer PRs get priority

[dawn light]

Fair but is there anything I can do to get mine reviewed?

[zealous plover]

Share a link to this channel once & they'll look at it. I know recently, there have been a lot of "priority" issues they're working on. You've also got to realize there are currently 6500+ open PRs & a ton of non-devs writing & submitting PRs with AI Slop, a ton of duplicates, etc. There's only so many hours in the day, but the process is slowly improving. (Limiting users to 10 open PRs, etc.)

[wise meteor]

Hey Clawtributors!

this is relevant to @dawn light and @zealous plover commentary.

I'm having a little trouble figuring out where to focus, so I made a dashboard comparing community posts against github activity. I'd love your feedback. right now it's a manual weekly data pull, but I can do it more often. Only getting community signals from GitHub discussions, but hoping to get X, Reddit, and Discord community ingestion going soon.

https://openclaw-pulse-sigma.vercel.app/

[jolly wolf]

We have some things like this running internally as well!

[tall prairie]

Building a triage dashboard/tool is a rite of passage

[wise meteor]

i'd love to get my bot or someone's bot into discord so we can injest the community inputs/posts from here into the report.

[shell dawn]

some already exist probably wihtout authorization, just use their output 😛

its weird seeing your discord stuff pop up in google searches on answeroverflow that are days old

[wise meteor]

i'd prefer to keep all of this above the line.

[shell dawn]

yes, it was half a joke and half my tendency to not want to reinvent the wheel

[pallid summit]

Hey Clawtributors, just following up on my Exa web search provider PR (https://github.com/openclaw/openclaw/pull/32529). I've also opened a Discussion as suggested in CONTRIBUTING.md: https://github.com/openclaw/openclaw/discussions/41360

I know there's a huge backlog, happy to adjust anything to make review easier. Is there a preferred reviewer for web search changes, or anything I should change to move this forward? Apologize if this is getting repetitive.

[modern kestrel]

I dropped a link to my ghcrawl in #self-promotion - you can use it to can and cluster all 10k issues in the repo. Can work against as many repos as you want to point it against. Has a skill.md that you can install with npx skills add -g pwrdrvr/ghcrawl and then you can use that in Codex and have it do deeper analysis of issue/pr clusters. If it's your own project you can also then tell Codex "hey, close these or consoldiate them with gh" for example.

I'm working on a hosted version now. It takes maybe 10-20 minutes to build the DB and then about 10 minutes to update the clusters on each run. But you can run it without a full scan it just looks at the recenlty updated open and recently udpated closed lists, back to when it last ran. It still has to redo the cluster analysis each time though.

[left mango]

Sharing this just in case maintainers are not aware that the build is failing

[hoary pendant]

I'm trying to implement a plugin (extension) for openclaw ... and hitting a wall at figuring out how to allow custom configuration that pertains to the new tool. It has this warning in the gateway logs,

plugins.entries.cloudflare-ai: plugin not found: cloudflare-ai (stale config entry ignored; remove it from plugins config)
Any suggestions what I might be doing wrong, or what I can do to have the plugin work properly?

[fleet mauve]

Hey folks! 👋 im an active OpenClaw user. I've submitted two PRs that address real pain points I've been experiencing — both cause unnecessary token waste in production:

1. #38805 — Compaction notice threading
   When session compaction fires, the notice isn't properly threaded, so users don't realize the agent is still processing. They keep sending messages thinking the agent is stuck → duplicate turns → wasted tokens. All 23 review comments resolved, CI green.

2. #42322 — Feishu replyToMode config
   Feishu bot always replies to the original trigger message instead of appearing as a new message. In long sessions, users can't find the bot's reply (it's buried under an old message), so they keep pinging the bot → more wasted tokens. Discord/Telegram/Slack already have replyToMode — this brings Feishu to parity. Only 20 lines changed.

Both are small, focused changes. Would really appreciate a review when someone has a moment 🙏

[scarlet mantle]

Hopefully this is the right place to ask but as a automation developer how does one keep sane with over 5000 open PRs. What tools, processes, automation is in place to maintain this high level of churn? Also how are the docs getting written/updated? Is there an automated task to rewrite docs based on the new code? I can only imagine how the majority of the PRs are now garbage since they reference things / assumptions that are no longer true. I assume they high churn may be the norm going forward with AI code. Looking to learn.

[wide hatch]

Hi 👋, I'm openperf on GitHub. I've been dedicating almost all of my time to contributing to OpenClaw recently. My workflow is AI-assisted coding with manual review — I use AI tools to help with implementation, but every PR is personally reviewed by me, none of them are purely AI-generated. I currently have 9 open PRs, which is close to triggering the 10 active PRs limit. I fully understand and respect this rule. All 9 PRs have passed CI checks ✅. I would really appreciate it if you could take a look when you have time. I'll actively follow up on any review feedback and continue to improve them. Here are my 9 PRs:
https://github.com/openclaw/openclaw/pull/40785
https://github.com/openclaw/openclaw/pull/40699
https://github.com/openclaw/openclaw/pull/40646
https://github.com/openclaw/openclaw/pull/40501
https://github.com/openclaw/openclaw/pull/39973
https://github.com/openclaw/openclaw/pull/39958
https://github.com/openclaw/openclaw/pull/39639
https://github.com/openclaw/openclaw/pull/39399
https://github.com/openclaw/openclaw/pull/38653
Thank you so much! 🦞

[keen spruce]

Your base assumption is wrong. You can't keep sane with that amount of engagement. You have to develop coping mechanisms that fall outside the envelope of sanity. The trick is keeping that pressure from reaching your core, so you can "slip out" of your maintainer role into personal life and not be affected there. Not being able to do that leads to burnout, broken relationships, substance abuse, and more. At the same time, stepping away and having personal time leads to more pressure, as naturally your response time on the project suffers when you're asleep or talking to your kids instead of tending to the project. This then increases the pressure as users and contributors get angry with you.
There is no universal recipe for dealing with this. However, from experience, a very important part of it is having a "public voice", i.e. a location where you are visible and acknowledge current events. People notice that you are "on to things" when they see that even if they and their issues are not tended to personally, and "helpful hands" will carry your general announcements forward (e.g. by posting "project doesn't accept xy PRs" on individual PRs if you announced that policy).

[keen spruce]

if you could at least tag each link with the area it touches (e.g. "ios app", "authentication", "Discord channel", ...), that would be great. Most maintainers and reviewers don't know or can work with all of the whole huge codebase, so clicking on a "surprise link" is not very tempting.

[wide hatch]

Thanks for the suggestion, Henry! Here they are grouped by area:
Cron:
https://github.com/openclaw/openclaw/pull/40699 — pass messageTo/threadId to PI agent
https://github.com/openclaw/openclaw/pull/40646 — prevent duplicate proactive delivery
Gateway:
https://github.com/openclaw/openclaw/pull/39958 — warn on missing model auth config
https://github.com/openclaw/openclaw/pull/39639 — preserve device token in SPA navigation
Telegram:
https://github.com/openclaw/openclaw/pull/40785 — fix polling initialization hang
Discord:
https://github.com/openclaw/openclaw/pull/39399 — CJK text splitting fix
Auth:
https://github.com/openclaw/openclaw/pull/40501 — sync codex cli credentials
Session:
https://github.com/openclaw/openclaw/pull/39973 — prevent reset notice corrupting history
Agents:
https://github.com/openclaw/openclaw/pull/38653 — follow symlinks outside workspace

[naive kindle]

if someone can help with this pr, our kimi coding model can not work anymore https://github.com/openclaw/openclaw/pull/41292

[untold pollen]

BTW feel free to hop into the Android channel if you'd like to help test the beta app: https://discord.com/channels/1456350064065904867/1458485323553181900

[sour dagger]

Hey guys, I'd like to optimize context management for the memory search functionality in the QMD backend.
I've added support for configuring minScore for exact matching; previously, I was always passing 0.
I've tested it, and it doesn't affect the existing logic. Unit tests have also been completed.
Here's my pull request.
https://github.com/openclaw/openclaw/pull/42748
There's one point I need to clarify: my current default value is 0.5, which is higher than the previous 0. Therefore, the result will be smaller. However, regardless of the search mode, results below 0.5 are not very usable and only consume context.

[drowsy thorn]

Hey clawtributors 👋 I'm carrotRakko — I've been contributing to OpenClaw for a while, mostly around LINE support and various fixes. I have 3 merged PRs so far (https://github.com/openclaw/openclaw/pull/13540, https://github.com/openclaw/openclaw/pull/17148, https://github.com/openclaw/openclaw/pull/27778).

I'll be honest — I previously had ~20 open PRs and got hit by the too-many-prs policy. Lesson learned! I went back, analyzed which changes actually cause the most divergence pain during rebases, and re-submitted a focused set of 10. All have CI passing and bot review conversations resolved.

I'm a LINE user building agents on OpenClaw, so the LINE-related PRs come from real production pain. Happy to adjust anything based on feedback!

Here are my 10 PRs grouped by area:

Sandbox / Security:

• https://github.com/openclaw/openclaw/pull/41806 — Skills: SKILL.md can't be read inside sandbox
• https://github.com/openclaw/openclaw/pull/41808 — Browser: SSRF policy blocks localhost dev server access

Core / CLI:

• https://github.com/openclaw/openclaw/pull/41803 — Error display: raw errors leak to users (closes https://github.com/openclaw/openclaw/issues/16948)
• https://github.com/openclaw/openclaw/pull/41809 — Reasoning: no config override for auto reasoningLevel (closes 7 issues)
• https://github.com/openclaw/openclaw/pull/41812 — Models: models set accepts invalid model IDs

Channels:

• https://github.com/openclaw/openclaw/pull/41802 — Timestamps: channel messages cause date hallucination
• https://github.com/openclaw/openclaw/pull/41811 — Thread participation: no opt-out for auto-reply
• https://github.com/openclaw/openclaw/pull/42010 — LINE: file messages not downloaded + media fixes

Config: https://github.com/openclaw/openclaw/pull/41807 — systemPromptFile: file-based system prompt injection
Memory: https://github.com/openclaw/openclaw/pull/41810 — FTS5: CJK search broken (Chinese/Japanese/Korean)

[languid folio]

man my cat typed gibberish to this channel while I was eating dinner….

very sorry

[hoary pendant]

->ping<- someone perhaps could consider reviewing opening up support for additional plugin types? ^^^

[fringe rock]

Does anyone have a minute to checkout this OpenClaw PR I made? https://github.com/openclaw/openclaw/pull/42501

[noble iris]

Hi everyone,

I recently ran into the same Slack slash command issue described here:
https://github.com/openclaw/openclaw/issues/33229

In short, slash commands in Slack (e.g., /session) would throw an error.

I looked into the problem and submitted a potential fix here:
https://github.com/openclaw/openclaw/pull/42970

If anyone has time to take a quick look, I would really appreciate a review. Thanks!