https://docs.all-hands.dev/ logo
Join Slack
Powered by
Feature Request: - we could add _the LLM profile i...
# proj-automation
c
Feature Request: • we could add the LLM profile in use, in the reviewer and QA agent comment (“This review is performed via Cloud Automation…”) • maybe mark these conversations “public share”, by default, if possible • or, add to the final comment, an archive with the logs. When they were in github actions, anyone could take a look at why the agent was mean 😅, or strange reasoning, or stuck etc. I think maybe that behavior is still useful - we just saw 2 instances running on the same PR, one with GPT-5.5 and another with Minimax, and both just stopped apparently
💡 1
@nutritious-balloon-65098 make an issue on openhands/automation on this. Investigate and propose solutions too
n
I'm on it! can track my progress here.
Done — I investigated and created the issue: github.com/OpenHands/automation/issues/221 I included: • The feature request context and observed symptoms. • Investigation notes from 
OpenHands/automation
, 
OpenHands/extensions
, and OpenHands conversation export APIs. • Phased solution proposals: a. Add LLM profile/resolved model and run links to reviewer/QA comments. b. Add first-class automation run artifacts/log archives. c. Support shareable conversation/run links with privacy controls. d. Add metadata/concurrency controls to distinguish duplicate or parallel runs. • Acceptance criteria and open questions. • The required AI disclosure line.
c
@nutritious-balloon-65098 OK, now read this comment, it restates this., and it also adds the last thing, make a new issue about that github.com/OpenHands/extensions/pull/378#…
n
Done — I read the linked review comment and created a separate issue for the additional point about controlling which LLM/profile reviewer and QA automations use: github.com/OpenHands/automation/issues/222 I kept it separate from the provenance/log visibility issue (#221) and framed it around: • deterministic default LLM profile selection, • per-run/profile override options, • keeping A/B or variant behavior explicitly opt-in, • persisting selected profile/model metadata on each run, • authorization and validation for overrides.
c
@nutritious-balloon-65098 thanks but your issue is too verbose, KISS! use the plain English skill
n
Done — I rewrote issue #222 in plain English and cut it down: github.com/OpenHands/automation/issues/222 It now focuses on: • the problem • what we should support • possible user flows • clear done criteria
c
Edit the comment I showed you, in some other PR, to now point to that issue too; and fix the Slack comment to do similarly
2 Views
Open in Slack
PreviousNext
Powered by