workflows: diff pull requests from the live base merge base - #31
Merged
Conversation
The code review diffed from pull_request.base.sha, which GitHub keeps at the base commit recorded when the head was last pushed. After the base branch moved and the pull request merged it back in, that stale commit was no longer an ancestor of every base change inside the head, and the incremental two-dot range from the last reviewed head carried the merged base commits as pull-request changes. The reviewer then reported the base branch's own work as out-of-scope implementation. The fetch step now also fetches the live base branch tip, and the diff listing starts at the merge base between that tip and the head, which is what the pull request's own "Files changed" view shows. Reuse and incremental decisions key on the merge base instead of the recorded base sha: a base branch that moves without being merged keeps reusing evidence, while merging it moves the merge base and starts a fresh full code review. pull_request.base.sha still selects the trusted reviewer checkout only. Validation - node .github/scripts/pr-review/test.mjs (new merge scenario fails against the previous prepare.mjs and passes now) - node .github/scripts/issue-review/test.mjs - node .github/scripts/pr-readiness/test.mjs - node .github/scripts/review-request/test.mjs - workflow YAML parses Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
❌ OpenAI PR Review: FAIL
Conclusion: Not ready. 1 readiness blocker must be resolved before merge.
Review checks
| Check | Result |
|---|---|
| PR format | ❌ FAIL |
| Issue design | ✅ PASS |
| Code & plan conformance | ✅ PASS |
Scope: eeedb45085 · b6904f156a..eeedb45085 · full · 1 diff chunk
Usage: 36s · 119,711 tokens · 76.1% cache hit · 3.214 credits
Blockers
pr-linkagePull request must natively close at least one same-repository Issue.
Summary
All completed code chunks were reviewed with no actionable findings or plan-conformance blockers. No linked Issue evidence was supplied.
Review metadata
- Commit:
eeedb45085 - Range:
b6904f156a..eeedb45085 - Mode:
full - Diff chunks:
1 - Model:
gpt-5.6-terra - Reasoning effort:
medium - Session:
repo-1309321116-pr-31-v2 - Generation:
d4e3d78ea48e188f8943b953d262186223d64a9d79478bf3a127dc56a0095ada - Evidence:
ef70e94d3fd285702f1312d7951004dd993c449185fefd86d7fa7b75145d94d9
Totals
- Input:
117,309 - Cached input:
89,214 - Cache write:
28,068 - Output:
2,402 - Reasoning:
1,364 - Total:
119,711 - Estimated credits:
3.214 - Credit rate per 1M tokens:
62.5input /6.25cached /375output
Token and cache details
| Stage | Mode | Target | Time | Input | Cached | Hit | Output | Total | Credits |
|---|---|---|---|---|---|---|---|---|---|
pr |
deterministic |
format rules |
0s | 0 | 0 | 0.0% | 0 | 0 | 0.000 |
pr |
full |
pr |
6s | 19,561 | 9,270 | 47.4% | 339 | 19,900 | 0.828 |
code |
full |
chunk 1/1 |
25s | 62,742 | 46,761 | 74.5% | 1,810 | 64,552 | 1.970 |
code |
full |
aggregate |
5s | 35,006 | 33,183 | 94.8% | 253 | 35,259 | 0.416 |
There was a problem hiding this comment.
✅ OpenAI PR Review: PASS
Conclusion: Ready from the OpenAI review perspective. PR format, linked Issue design, and code/plan conformance passed with no actionable findings.
Review checks
| Check | Result |
|---|---|
| PR format | ✅ PASS |
| Issue design | ✅ PASS |
| Code & plan conformance | ✅ PASS |
Scope: eeedb45085 · eeedb45085..eeedb45085 · incremental · 1 diff chunk
Usage: 30s · 192,671 tokens · 75.0% cache hit · 4.537 credits
Summary
The incremental generation changes no code and retains the prior full-review result: no actionable findings or plan-conformance blockers. Linked Issue #32 provides a reviewed, implementation-ready plan.
Review metadata
- Commit:
eeedb45085 - Range:
eeedb45085..eeedb45085 - Mode:
incremental - Diff chunks:
1 - Model:
gpt-5.6-terra - Reasoning effort:
medium - Session:
repo-1309321116-pr-31-v2 - Generation:
099cffec34e4910239347104b7f21c2f8e0a6417224ba65499b5fbac6d1e0ddf - Evidence:
592e3281be52ce5fe705f894267314e0f2583f66f8282d1395f263946afc065f
Totals
- Input:
190,913 - Cached input:
143,195 - Cache write:
47,691 - Output:
1,758 - Reasoning:
600 - Total:
192,671 - Estimated credits:
4.537 - Credit rate per 1M tokens:
62.5input /6.25cached /375output
Token and cache details
| Stage | Mode | Target | Time | Input | Cached | Hit | Output | Total | Credits |
|---|---|---|---|---|---|---|---|---|---|
pr |
deterministic |
format rules |
0s | 0 | 0 | 0.0% | 0 | 0 | 0.000 |
pr |
incremental |
pr |
7s | 36,569 | 17,969 | 49.1% | 304 | 36,873 | 1.389 |
issue |
full |
Issue #32 |
15s | 81,792 | 78,753 | 96.3% | 952 | 82,744 | 1.039 |
code |
reused |
chunk 1 |
0s | 0 | 0 | 0.0% | 0 | 0 | 0.000 |
code |
incremental |
aggregate |
8s | 72,552 | 46,473 | 64.1% | 502 | 73,054 | 2.109 |
idy
added a commit
that referenced
this pull request
Sep 7, 2026
…ew (#34) A full generation re-reviews the complete current diff from its merge base, which since #31 happens whenever the base branch is merged into the pull request. The aggregation still received the previous code review and was told to preserve still-applicable findings, so blockers about an earlier range survived diffs that no longer contained their subject. Only an incremental generation now carries the previous result, and the full-mode prompt says nothing is carried over. Validation - node .github/scripts/pr-review/test.mjs (the full-generation assertion fails against the previous run.mjs and passes now) - node .github/scripts/issue-review/test.mjs - node .github/scripts/pr-readiness/test.mjs - node .github/scripts/review-request/test.mjs Closes #33 Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
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
The code review stage diffed from
pull_request.base.sha. GitHub keeps that value at the base commit recorded when the head was last pushed, so once the base branch moves and the pull request merges it back in, the stale base is no longer an ancestor of every base change inside the head. The incremental two-dot range from the last reviewed head then carried the merged base commits as pull-request changes, and the reviewer reported the base branch's own work as out-of-scope implementation. This happened on GizClaw/gizclaw#1160 aftermainwas merged in to resolve a conflict: the review flagged main's Doubao and Flowcraft commits while GitHub's own diff showed only the PR's files.refs/heads/<base_ref>) and exportsPR_BASE_TIP_SHA.prepare.mjsstarts the full diff and the effective added-line ranges at the merge base between that tip and the head, which is exactly the PR's "Files changed".merge_base_shainstead ofbase_sha. A base branch that moves without being merged keeps the merge base and reuses evidence; merging it moves the merge base and starts a fresh full code review, so base-branch commits never enter a reviewed range.pull_request.base.shastill selects the trusted reviewer checkout only.PR_BASE_TIP_SHAfalls back to it when unset, so existing callers ofprepare.mjskeep working.Validation
node .github/scripts/pr-review/test.mjsprepare.mjsand passes nownode .github/scripts/issue-review/test.mjsnode .github/scripts/pr-readiness/test.mjsnode .github/scripts/review-request/test.mjsNot run: a live workflow run against a real pull request. The
base_refoutput andPR_BASE_TIP_SHAexport follow the existing fetch step's structure, and the script path is covered by the unit scenario.Not run cleanly:
actionlintreports pre-existingjob.workflow_sha/job.workflow_repositoryexpression warnings on lines this PR does not touch; no new diagnostics.Closes #32
🤖 Generated with Claude Code