related to this: is there a low- or no-code method of implementing the 4 roles in the software factory pattern which appears to be hard-coded for github?

Yes, the software-factory is just for github for instance

Yeah, and specifically the way you could do this is tell the agent: “please implement a software factory for me, but do it for forgejo”. It will read the github instructions, but probably be able to adapt it for forgejo. Please tell me how it goes. If it goes well we’d definitely be happy to have a contribution of the automations that it creates to the OpenHands/extensions repo. If it doesn’t go well we’d be happy to help make it go better

Looking at the tarballs for the extensions and the low-level python code (none of which I understand of course), it definitely did not occur to me to simply instruct the chat to re-implement the pattern.

Cool, tell me if it works!

i don't have software projects or issues set up yet, but it looks like it tried to do the right thing.

Awesome, excited to see it working 🚀

For others seeing this thread, and FWIW, my setup is ghcr.io/openhands/agent-canvas:latest running on Mac Studio M5 Max 64GB, along with qwen3.6-35b-a3b via LM Studio. Web UI on a Mac Mini M4 16GB.

You could ask the agent to debug?

suggest a prompt for me?

"The issue-to-PR automation didn't run when I added openhands label to the issue. Why not? Find it and investigate"

i'm using qwen3-coder-30b-a3b-instruct-mlx and it's slow af. it's not the model tho, it responds instantly using the LM studio chat interactively. Looking at the LM studio logs, it seems like agent canvas is pushing 16k worth of context every single turn. Also, agent canvas says "Thinking" for like 30 seconds, but the coder model doesn't even support reasoning (yes the MoE model does, but I don't know the magic incantations to turn it off). So why does the "Thinking" bubble show while the system stalls for 30-60 seconds? Back on topic: i wrote a more detailed prompt than that, asking for root cause without making code changes, Numerous corrective prompts were required to fix invalid conclusions and wrong assumptions. End result was this:

Current State Analysis
Root Cause: One or more Forgejo API calls in the automation scripts silently fail during state transitions.
Specific Issue:
Triage runs successfully (posts comment) but fails to properly remove "openhands" label
This causes continuous triage loop
Issue-to-PR never executes because trigger condition isn't properly cleared
Why it happens:
Forgejo API operations (label removal, issue body updates) return success but don't actually complete the operation in your specific environment.
Behavioral Proof:
The exact pattern you described - triage working but issue-to-PR not executing at all, with continuous looping.
Simple Answer: The automation scripts have runtime failures in specific Forgejo environments that prevent proper state transitions.

Root Cause - Specific Modules and Lines
Module 1: forgejo-issue-triage/main.py
Line 337: remove_label(owner, repo, issue_num, TRIGGER_LABEL)
Line 352: forgejo_api("PATCH", f"/api/v1/repos/{owner}/{repo}/issues/{issue_num}", {"body": new_body})
Module 2: forgejo-issue-to-pr/main.py
Line 261: create_branch(owner, repo, default_branch, branch_name)
Line 272-281: create_commit() operations
Line 291: create_pull_request()
Specific Issue:
The remove_label() call in triage (line 337) and the issue body PATCH operation (line 352) are most likely to silently fail in your Forgejo environment, causing the state transition failure that creates the infinite loop behavior.

What is meant by "in your Forgejo environment" and " in specific Forgejo environments" and "in your specific environment" here? Prior testing (hi chat, go process the issue on the test3 repo) verified the system works with forgejo, read an issue, and followed the issue instructions to modify a README on a branch and create a PR against it for human approval/merge. I'm just a noob, so a lot of this is obtuse and opaque to me. I would appreciate if someone knowledgable could repro with forgejo 17 and fix the 4 automations. Also would like advice on how to improve performance.