Skip to content

feat(gemini-review): review fork PRs, review only; v0.5.0 - #24

Merged
h0ffmann merged 2 commits into
mainfrom
claude/project-thread-g5dwen
Oct 4, 2026
Merged

h0ffmann merged 2 commits into
mainfrom
claude/project-thread-g5dwen

Conversation

@h0ffmann

@h0ffmann h0ffmann commented Oct 4, 2026 •

Copy link
Copy Markdown
Contributor

Requested by Hoffmann · project thread

Summary: Before: requesting gemini on a fork PR (marola-dev/marola-site#54) did nothing, because the job skipped forks. After: a fork PR gets the Gemini review, with suggestion blocks, but no fix commit. Same-repo PRs work as before.

MIP none. Spec: #23, from marola-dev/marola#641's "a fork PR gets a review but no push"
Tested ✅ gates · ⬜ e2e · ⬜ live · ⬜ ci-only: gemini_review.py --self-test (the --root case failed first), tests/self-tests.sh, actionlint, ruff and docs-lint all pass. Live check: re-request gemini on marola-site#54 once its caller is on v0.5.0
Cost ~$0.6 est.

What changed

  • gemini-review.yml no longer skips forks. On a fork PR, the workspace is the base commit, so the guide is the repo's own. The fork's head is checked out into .pr-head as text only, nothing from it runs, and the fix step is skipped.
  • gemini_review.py review --root DIR is the tree used to anchor suggestions (default .).
  • Callers trigger on pull_request_target so a fork run gets the secrets. This is safe because only someone with triage access can request the team. This repo's own gemini.yml moves too.
  • devkit-ci and api-docs drop magic-nix-cache-action, as ci: drop magic-nix-cache from quality-other and repo-stats marola#658 did. Its post step spent 2m40s of a 6m23s devkit CI run uploading paths cache.nixos.org already serves.
  • README, docs/4-reference_workflows.md, the CHANGELOG and every pin move to v0.5.0.

How: After merging, tag v0.5.0. The six caller PRs then switch to pull_request_target and @v0.5.0.

Closes #23

https://claude.ai/code/session_01NYzPmf2qCK2s2RYccvBVBY


Generated by Claude Code


Generated by Claude Code

A fork PR (marola-site#54) never got a review: the job skipped forks because
pull_request gives them no secrets, though marola#641 asked for "a review but
no push". Callers now trigger on pull_request_target; on a fork the workspace
is the base branch, the fork's head is read as text from .pr-head
(review --root) and the fix step is skipped. Same-repo PRs are unchanged.

Closes #23

Tested: gemini_review.py --self-test ok (--root case red first); tests/self-tests.sh all ok; actionlint ok; docs-lint clean; ruff ok
Cost: ~$0.6 est.
Co-Authored-By: Claude <noreply@anthropic.com>
@h0ffmann
h0ffmann marked this pull request as ready for review October 4, 2026 22:03
Its post step spent 2m40s of a 6m23s devkit CI run (main, 2026-10-04)
uploading store paths cache.nixos.org already serves; marola dropped it
for the same reason in marola-dev/marola#658.

Refs marola-dev/marola#657

Tested: actionlint ok; docs-lint clean; step timings from devkit CI on main
Cost: ~$0.2 est.
Co-Authored-By: Claude <noreply@anthropic.com>
@h0ffmann
h0ffmann merged commit 6947ba1 into main Oct 4, 2026
10 of 11 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

gemini-review: review fork PRs (review only, no push)

2 participants