Bump both reviewer pins to ai-review-prompts@ea19009#88
Merged
Conversation
Adopts the promoted Gemini calibration (ai-review-prompts#41) and restores both reviewers to a single shared pin. Gemini changes vs the prior pin (128656e40): - Single-shot architecture validated empirically over 5 iteration rounds in PR #87 (gemini/inline-calibrate) - Prior-review body fetched and injected as continuity context when the prior had findings (### N. headers); skipped on no-blockers priors so the model doesn't get primed to hunt for issues against uninformative context - Prompt's "your output IS the PR comment" section reworded to allow shell/file research tools while still forbidding self- posting (the workflow handles posting) - post-review-comment.sh now strips leading whitespace from the marker line so a non-whitespace-stable provider can't break the marker check Claude reviewer is byte-identical between f22bf7d and ea19009 in ai-review-prompts; bumping the pin is a no-op for Claude behavior but restores the two reviewers to a single shared upgrade motion. The "intentionally NOT in lockstep" comment from the divergence period is replaced with a note that lockstep is restored. PR #87 (gemini/inline-calibrate) is now superseded by this PR and will be closed without merging. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
|
Reviewed; no blockers found. |
Mirrors the hardening on claude-review.yml — explicit grant at the calling-job level for the union the reusable's authorize + review jobs need (contents: read + pull-requests: write + id-token: write). The workflow happens to work without this on oauth today because the repo default is permissive enough, but the explicit grant survives repo-default changes and matches claude-review.yml's shape and discipline. Surfaced as a finding by Gemini's own review of this PR (#88) — exactly the loop the calibration was built for. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This was referenced May 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adopts the promoted Gemini calibration (ai-review-prompts#41) and restores both reviewers to a single shared pin (
ea190091328bcee674c4739ccc97dda177ecf0c5).Gemini changes vs the prior pin (
128656e40)gemini/inline-calibrate). The ai-review-prompts main branch had drifted onto an MCP-based rewrite that nothing in production was using; fix: default redirectUri matches Harper dev config (http://localhost:9926) #41 replaces it with the calibrated single-shot.### N.-shaped findings, and injects it into the prompt as a "Continuity reference" section. Skipped on no-blockers priors so the model isn't primed to hunt for issues against uninformative context.gemini-clihas them by default and the model does use them to research the diff) while still forbidding self-posting (the workflow handles posting). Resolves a latent self-contradicting prompt that surfaced as a finding once the calibration loop got fast enough to see it.post-review-comment.shmarker check now strips leading whitespace from the first non-blank line so a non-whitespace-stable provider can't break the marker check.Claude lockstep restored
_claude-review.ymlis byte-identical betweenf22bf7dandea19009in ai-review-prompts (the divergence period only touched Gemini files). Bumping Claude's pin in lockstep is a no-op for Claude behavior but restores both reviewers to a single shared upgrade motion — the "intentionally NOT in lockstep" comment from the divergence period is replaced with a note that lockstep is restored.Follow-ups
gemini/inline-calibrate) is now superseded by this PR — will close without merging once this lands.Test plan
claude-reviewandgemini-reviewworkflows fire on this PR🤖 Generated with Claude Code