fix: honor selected LLM profiles in OpenHands/extensions #547

[enyst]

• A human has tested these changes.

---

I am an AI agent acting on behalf of Engel.

Why

OpenHands#547 currently ignores AUTOMATION_MODEL: choosing gpt-latest-med can launch the user's unrelated default LLM. This follow-up targets OpenHands#547's branch at 1b6a4a3bbfdfa4103b8bb0c32b8754a3afbd9b56.

Summary

• Resolve the selected profile for each new GitHub or Slack conversation, preserve its complete LLM configuration, and use the same resolved profile/model for provenance.
• Use concrete default settings only when the selection is absent or returns 404; keep other errors visible and clearly reject unresolved provider-linked credentials.
• Add HTTP regressions, document behavior, and regenerate reviewer bundle 1.0.3.

Human explanation: Read the rendered .pr document · source.
It explains the original bug, selection/fallback rules, existing credential handling, standalone packaging, evidence, and rollout.

Issue Number

Follow-up to OpenHands#547; related to OpenHands#548 and OpenHands/automation#430.

How to Test

uv sync --group test
uv run pytest tests/test_automation_llm_provenance.py tests/test_slack_channel_monitor.py skills/github-pr-reviewer/tests/test_main.py -q
uv run pytest tests skills/github-pr-reviewer/tests -q
npm run build:automations
npm run build:skills
uv run python scripts/sync_extensions.py --check
git diff --check

99 focused tests and 896 full-suite tests pass; 24 skipped.
All four new selected-profile HTTP cases fail on the original PR head and pass here.
They exercise the actual profile GET and conversation POST for both scripts, checking credentials, base URL, reasoning effort, selected model, and provenance.
The local HTTP service uses synthetic credentials and models the linked-profile response; it does not replace #4952's server-side tests or demonstrate production deployment.
Syntax checks, bundle/source equality, repository sync, and HTML structure checks pass.

Notes

GitHub rejected creating this directly against OpenHands/extensions with 403 (resource not accessible by integration). This PR targets an exact mirror of OpenHands#547's current head in the enyst fork, keeping the diff limited to this follow-up. Merging here will not update OpenHands#547; apply commit 7a8b3789b879d717dd893b0840b4c28fb74a01ee to OpenHands/extensions:openhands/show-llm-provenance.

Provider-linked profiles require Agent Server #4952 (or equivalent) to be deployed. Inline profiles use the existing server API.
The script uses the existing authenticated plaintext read and sends the configuration back to the same Agent Server; it does not activate profiles globally.
A server-side profile selector remains a separate API improvement, explained in the .pr document.
Default provenance uses default, since the separate active-profile pointer may not identify the concrete settings actually used.

The repository sync check retains the existing non-blocking issue-duplicate-checker coverage warning.
The generic Codex skill validator rejects the existing OpenHands triggers extension; that frontmatter is unchanged and supported by the repository's own validator.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}