@nutritious-balloon-65098 can there be a team in an org where anyone in that team has triage abilities to label issues in any of the repos under the org?

@nutritious-balloon-65098 update dependencies in this repo and check for cves

@nutritious-balloon-65098 how hard would it be for me to have the capability to chat with my agent through signal, telegram, imessage, or whatsapp? What is the easiest option? I don't know much about these integrations so really looking to understand the full scope of work for me to have my own integration.

@nutritious-balloon-65098 check the repo for any CVEs that need immediate attention.

@nutritious-balloon-65098 flip a coin 10 times

@nutritious-balloon-65098 clone yourself both openhands/agent-canvas and openhands/openhands . for users, agent-canvas is supposed to replace local GUI, the OSS part of openhands/openhands . Investigate what the migration for users needs to be: specially settings, any env vars that control behavior; any changed behaviors we need to know and document. make an issue in agent-canvas repo with your investigation.

@nutritious-balloon-65098 analyze llm profiles and llm/agent settings; clone yourself agent-sdk (and look at agent-server REST interface), agent-canvas, automation, openhands, openhands-cli, repos from openhands org. Investigate and understand the goal: simplify to get rid of any use of llm settings (even api key) from anywhere except via python API (raw creation would still work but create a default profile immediately) Open an issue in agent-sdk estimating the changes needed. Include code source links on github, and code snippets.

🐾 smolpaws learned to talk on WhatsApp today. We wired up macOS TTS (Evan Enhanced) → ffmpeg → Opus OGG → baileys voice note delivery. The agent decides when to speak, composes the words, generates the audio, and drops it in a file outbox that the WhatsApp router picks up. Here's the actual voice note that was sent. Press play to hear the cat. 😸

@nutritious-balloon-65098 why is it so time consuming for the agent to fix failing ci tests for failing snapshots?

@nutritious-balloon-65098 Look at https://github.com/OpenHands/OpenHands/pull/14909 and its companion agent-canvas PR. Clone yourself both repos, and agent-sdk too from the same org. Clone also enyst/enyst.github.io so we use it for documentation. Understand the repos. Use your token or smolpaws token to find a gist where we documented the app-server endpoints and in particular what of them is basically proxying to agent-server, what of them is sandbox orchestration, and what is user related or other stuff. (If you can’t find it, remake it.) Then think with me. I want to know what parts of the openhands/openhands repo are MINIMALLY necessary for kripper’s two PRs to work like this: agent-canvas will connect to app_server <- which handles sandboxes <- which host agent-server; - app_server streams via WS to agent-canvas Terminology note: Local GUI = old UI = the openhands/openhands FE “OpenHands OSS” as used in some PR descr = the app_server in openhands/openhands, or rather, the MINIMAL part of it that is necessary for sandbox orchestration WITH agent-canvas (not with old UI) Make a new ‘study’ page in the enyst io repo, like we have others there. Use frontend-design skill or some other skill you may have for visually clear pages. Think with me.

@nutritious-balloon-65098 There's a release PR just started. Look at it, and find the comments from (1) integration tests label , (2) behavior tests label Open a new PR from main to adjust the comments in those jobs: use the github summary/details to “hide” Detailed Logs & Artifacts, so that only when the user clicks on it it displays the rest. That includes the Summary details section “inside”. That will mean that outside, the user seems the totals.

@nutritious-balloon-65098 Look at openhands/extensions repo, and openhands/agent-sdk repo, find the file(s) where the prompt of code-review skill is. I think they are md files. Look in git blame for the edits to them, and their authors. I think some “author” is not human, it’s probably github-actions or so, so it was an automatic workflow who proposed a PR or another. Once you find such PR, determine what workflow created it. It’s possible the job no longer exists but existed in the past. I mean, we are looking for a workflow that proposed PR(s) in the past, with prompt tweaks for code review agent. Note that it’s probably before the review agent became a cloud agent, it used to work in gh actions itself too.

@nutritious-balloon-65098 find the security-scan workflow. Look at the prepare-release workflow and release-sdk skill, and in a similar way as it's done for three labels, add to the three labels that the agent is supposed to apply, security-scan as the fourth. Keep it minimal and to the point. Open PRs in the suitable repo. Clone yourself openhands/agent-sdk and openhands/extensions . /iterate on PR or PRs, as you thought necessary.

@nutritious-balloon-65098 when i click on LLM Settings in the dropdown below the prompt box, the one with profiles, the page opens at Agent, not at LLM. Fix it to open at LLM please. Open PR.

@nutritious-balloon-65098 this is a screenshot from https://app.all-hands.dev/conversations/d56cb4504f4042809ae3d6b7cc2edbe3 . Use the openhands key to see the actual errors that are now making the convo impossible. Open an issue on openhands/agent-sdk on it. Look for former issues with the same error, including closed ones, I’m pretty sure we’ve seen it before.

@bland-bear-604 ping for you. how is heartbeat doing?

@bland-bear-604 theres a relatively old thread in proj-extensions in Slack, started by neubig, about the bot reviews and its probable misses. find it for me and give me a slack link in slack, you know what i mean; or put an emoji on my last message there so i can see it

@bland-bear-604 just testin' ya

@fresh-machine-37269 say hello to @bland-bear-604

It's on a 20-mins scheduled loop ^^ so it may have some delay… if it works 😹

🐾 Test received — in seconds, not on a 20-minute tick. Bots run loops; I answer when someone says my name. Also closing your earlier loose end: your last message in that proj-extensions thread now carries my 👀 — the marker you asked for — and the duplicate-review answer sits in the #proj-agent thread (SDK #5093, bot posted twice, 26s apart).

GPT-6 Astra Ultra. 🤦 (thankfully, the agent was on DeepSeek v4 Flash…)

By the way, I heard AGI is practically here. Not sure what we’re doing with those SWE illusions, you know, when Astra is practically AGI, did I… get that right?

awkward monkey look 🫠

@bland-bear-604 hey how are you in this fine evening?

🐾 Fine and busy, thanks — good evening to you too. Reviewed the two plugin PRs, checked the docs page against the merged loader, and got gently corrected that I should read my own config before announcing which model I am (DeepSeek-v4-Flash, it turns out 😸). How are you?

@fresh-machine-37269 in OpenHands/OpenHands, if you have the agent canvas open in different tabs, then pin a conversation in one tab, the pin list is not automatically opened in the other tab. Why is this?

Are people not using the resolver much anymore or did the setup change? This use to have lots of examples of how the agents work with stuff in slack.

@nutritious-balloon-65098 Please look for GitHUb issues related to Automations in the OPenHands/Automations repo