fix(dev-lead): repair auto-rebase conflict detection and applied-rate signal by don-petry · Pull Request #812 · combsmith/.github-private

don-petry · 2026-06-20T00:35:53Z

Why

The #737 auto-rebase health report (tracking issue #767) shows 0% applied every day despite 50–86 conflict sentinels/week and ~70% "responses". Investigation (discussion #735) found the agentic conflict-resolution path is both mis-instrumented and carrying a real conflict-detection bug — so "0% applied" was not a clean signal that resolution fails, but it was hiding genuine breakage.

Root cause

Bogus conflict-file detection. scripts/dev-lead-fix-reviews.sh used the deprecated 3-arg git merge-tree <base> HEAD <base> and parsed it with grep "^changed in both" | awk '{print $NF}'. The last field of the changed in both section header is the literal word both — the filenames are on the indented our/their lines. So CONFLICTING_FILES was set to both (a nonexistent path) on every conflict.
Prompt then weaponized the garbage. prompts/dev-lead/rebase.md injected that list and constrained the engine to "only modify the conflicting files listed above" — i.e. only both — so the engine couldn't touch the actually-conflicted files and aborted → no-changes.
applied was structurally unreachable. The prompt has the engine force-push the rebased branch itself (step 6), so on success commit_and_push finds nothing to push and posts no-changes. status=applied only fired if the engine committed but forgot to push — so even successful rebases were recorded as no-changes, making the metric ~0 by construction.

Changes

scripts/dev-lead-fix-reviews.sh
- New detect_conflicting_paths <base_ref> helper: version-robust trial merge (git merge --no-commit --no-ff → git diff --name-only --diff-filter=U → git merge --abort).
- Rebase path uses it for CONFLICTING_FILES, and after a successful run re-checks the base conflict state to record applied (no conflicts remain) vs no-changes (conflicts persist) instead of relying on commit_and_push seeing leftover work.
prompts/dev-lead/rebase.md
- Conflicting-file list is now advisory (may be empty/incomplete/stale); the engine is told to resolve whatever git rebase actually flags and not to skip a real conflict just because it isn't listed.

Verification

shellcheck --severity=warning -x scripts/dev-lead-fix-reviews.sh — clean.
bats tests/dev-lead/unit/test_fix_reviews.bats (86), test_intent_reviews.bats, test_auto_rebase_retry.bats — pass.
tests/dev-lead/integration/test_prompt_coverage.sh — pass (rebase.md variable declarations intact).

Note for the initiative (epic #736 / story #739)

This corrects the premise behind the earlier "0% applied" reading: that number was not evidence that agentic conflict-resolution is non-functional — it was a measurement artifact plus the detection bug. Story #739's decision record should source resolution-rate numbers from the post-fix report window, not the broken metric.

🤖 Generated with Claude Code

Generated by Claude Code

… signal The auto-rebase agentic conflict-resolution path was effectively broken and mis-instrumented, surfacing as 0% "applied" in the #737 health report (discussion #735, tracking issue #767). Three fixes: 1. Conflict detection (scripts/dev-lead-fix-reviews.sh): the old `git merge-tree <base> HEAD <base>` 3-arg form prints a "changed in both" section header whose last field is the literal word "both" (the filename is on the indented our/their lines), so `awk '{print $NF}'` fed a bogus "both" path into the rebase prompt. Replace with a version-robust trial merge (`git merge --no-commit` → `git diff --diff-filter=U` → `git merge --abort`) via a new detect_conflicting_paths helper. 2. Prompt hardening (prompts/dev-lead/rebase.md): make the conflicting-file list advisory ("may be empty/incomplete/stale") and instruct the engine to resolve whatever `git rebase` actually flags, so a bad/empty hint can no longer silently block resolution. 3. Applied-rate signal (rebase path): the prompt has the engine force-push the rebased branch itself, so commit_and_push finds nothing to push on success and every successful rebase was recorded as "no-changes" (0% applied). After a successful run, re-check the base conflict state — if no conflicts remain, record status=applied; otherwise no-changes. Verified: shellcheck --severity=warning -x clean; dev-lead bats unit suites and prompt-coverage pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QrRBgQJ7QqjMsRySzkPqwJ

chatgpt-codex-connector · 2026-06-20T00:35:56Z

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

coderabbitai · 2026-06-20T00:36:00Z

Warning

Review limit reached

@don-petry, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 46 minutes and 21 seconds. Learn how PR review limits work.

Your organization has used up its prepaid credits, and credit purchases are no longer available. Enable the review add-on in the billing tab to keep reviews running — you're only billed for reviews past your plan's rate limits ($0.25/file).

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based credits.

🚦 How do rate limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan refill rate.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, the refill rate gradually slows as usage increases. The highest same-day bursts are limited more strictly.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

Run ID: bfe7f7a2-7190-4b3d-9f5f-398f451a687c

📥 Commits

Reviewing files that changed from the base of the PR and between 9cb66bb and cbe4442.

📒 Files selected for processing (2)

prompts/dev-lead/rebase.md
scripts/dev-lead-fix-reviews.sh

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch claude/github-discussion-735-yvc0iq

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

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

don-petry · 2026-06-20T00:36:39Z

Dev-Lead — review-changes (no-changes)

No changes were needed for this PR.

donpetry-bot · 2026-06-20T00:37:14Z

Advisory bots were rate-limited; auto-approval is withheld until they recover. pr-review-sweep will re-review this PR after 2026-06-20T01:37:13Z.

don-petry · 2026-06-20T00:37:29Z

Dev-Lead — fix-bot-comment (no-changes)

Agent reasoning

Issues addressed: 0
Files changed: None
Skipped (informational): 0
Status: No action required — Quality Gate passed with zero issues.
```

gemini-code-assist

Code Review

This pull request improves the merge conflict detection and rebase handling in the dev-lead agent. It replaces a fragile git merge-tree parser with a robust trial-merge helper (detect_conflicting_paths) and updates the rebase prompt to clarify that the conflicting files list is advisory. Additionally, it fixes a telemetry bug where successful force-pushed rebases were incorrectly reported as having "no changes" by re-checking the conflict state after the run. Feedback was provided to add a defensive guard clause in the new helper function to handle empty arguments.

don-petry · 2026-06-20T00:43:46Z

Dev-Lead — fix-reviews (applied)

Changes committed and pushed.

sonarqubecloud · 2026-06-20T00:45:11Z

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

don-petry · 2026-06-20T00:46:47Z

Dev-Lead — review-changes (no-changes)

No changes were needed for this PR.

donpetry-bot

Automated review — APPROVED ✓

Risk: LOW
Reviewed commit: cbe4442a705e79bde8885c88d387ccdf46b465a1
Review mode: triage-approved (single reviewer)

Summary

Two-file fix (34+/5-) to the dev-lead auto-rebase path. Replaces a broken 3-arg git merge-tree | grep 'changed in both' | awk '{print $NF}' parser (which always yielded the literal path "both") with a version-robust trial-merge helper detect_conflicting_paths, and fixes a telemetry artifact where engine-self-force-pushed rebases were always recorded as no-changes (the "0% applied" signal) by re-checking base conflict state to emit applied vs no-changes. The rebase prompt's conflicting-file list is downgraded to advisory. Triage cleared this as low-risk; confirmation review agrees.

Linked issue analysis

No formal closing-issue reference, but the PR is well-anchored to discussion #735 and tracking issues #737/#767/#739. The change directly addresses the documented root cause (bogus both path + structurally-unreachable applied status). The PR body also correctly flags that the prior '0% applied' metric was a measurement artifact, not proof that resolution fails — a useful correction for story #739's decision record.

Findings

No blocking issues.

detect_conflicting_paths guards empty input ([[ -z "$base" ]] && return 0), redirects merge noise, and aborts the trial merge so the worktree is left clean. This satisfies Gemini's only review note (defensive guard), which was addressed in the follow-up commit.
All git refs are quoted; $BASE_REF/$CONFLICTING_FILES come from PR metadata and are not interpolated into a shell-eval context — no injection surface. No secrets, auth, crypto, or migrations touched. shellcheck + bats (86) + prompt-coverage all pass in CI.
Non-blocking observation: the post-run applied re-check merges origin/$BASE_REF into the local HEAD. If the engine force-pushes the rebased branch without the local checkout advancing, a successful rebase could still read as conflicts-persist → no-changes. The relevant bats suites pass, so this appears handled, but it is the main correctness assumption worth keeping an eye on in the post-fix telemetry window.

CI status

All required checks green (shellcheck/ShellCheck, bats, unit-tests, unit, prompt-coverage, CodeQL, SonarCloud, agent-shield, Agent Security Scan, gitleaks secret scan, validate-* etc.); dependency-audit/dependabot/ci-relay jobs SKIPPED as expected. No failures. mergeStateStatus=BLOCKED is the pending human/required-review gate, not a check failure. Note: MCP secret scan unavailable (no GitHub Advanced Security); gitleaks CI check covers secret detection and passed.

Reviewed automatically by the PR-review agent (single-reviewer mode: fable 5). Reply if you need a human review.

`.github-private` is ring 0 in the agentic release strategy — the self-host canary that runs a release before it is centrally promoted to `dev-lead/stable` for consumer repos. The caller was pinned to `@dev-lead/stable`, so this repo was running the same channel as consumers and never actually exercised new releases first. Point the `uses:` ref and `agent_ref` at `@dev-lead/next` so self-host runs the canary channel. Concretely this picks up dev-lead v1.4.0, which contains the auto-rebase conflict-detection + applied-rate fixes (#812) — so the rebase path and the #737/#767 instrumentation start reflecting the corrected behavior here before stable promotion. Stub validator (tests/dev-lead/integration/test_dev_lead_stub.sh) passes — it accepts any first-party channel tag. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QrRBgQJ7QqjMsRySzkPqwJ

…859) `.github-private` is ring 0 in the agentic release strategy — the self-host canary that runs a release before it is centrally promoted to `dev-lead/stable` for consumer repos. The caller was pinned to `@dev-lead/stable`, so this repo was running the same channel as consumers and never actually exercised new releases first. Point the `uses:` ref and `agent_ref` at `@dev-lead/next` so self-host runs the canary channel. Concretely this picks up dev-lead v1.4.0, which contains the auto-rebase conflict-detection + applied-rate fixes (#812) — so the rebase path and the #737/#767 instrumentation start reflecting the corrected behavior here before stable promotion. Stub validator (tests/dev-lead/integration/test_dev_lead_stub.sh) passes — it accepts any first-party channel tag. Claude-Session: https://claude.ai/code/session_01QrRBgQJ7QqjMsRySzkPqwJ Co-authored-by: donpetry-bot <{}+donpetry-bot@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… signal (#812) * fix(dev-lead): repair auto-rebase conflict detection and applied-rate signal The auto-rebase agentic conflict-resolution path was effectively broken and mis-instrumented, surfacing as 0% "applied" in the #737 health report (discussion #735, tracking issue #767). Three fixes: 1. Conflict detection (scripts/dev-lead-fix-reviews.sh): the old `git merge-tree <base> HEAD <base>` 3-arg form prints a "changed in both" section header whose last field is the literal word "both" (the filename is on the indented our/their lines), so `awk '{print $NF}'` fed a bogus "both" path into the rebase prompt. Replace with a version-robust trial merge (`git merge --no-commit` → `git diff --diff-filter=U` → `git merge --abort`) via a new detect_conflicting_paths helper. 2. Prompt hardening (prompts/dev-lead/rebase.md): make the conflicting-file list advisory ("may be empty/incomplete/stale") and instruct the engine to resolve whatever `git rebase` actually flags, so a bad/empty hint can no longer silently block resolution. 3. Applied-rate signal (rebase path): the prompt has the engine force-push the rebased branch itself, so commit_and_push finds nothing to push on success and every successful rebase was recorded as "no-changes" (0% applied). After a successful run, re-check the base conflict state — if no conflicts remain, record status=applied; otherwise no-changes. Verified: shellcheck --severity=warning -x clean; dev-lead bats unit suites and prompt-coverage pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01QrRBgQJ7QqjMsRySzkPqwJ * fix(reviews): address review comments [skip ci-relay] --------- Co-authored-by: donpetry-bot <{}+donpetry-bot@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: donpetry-bot <281750570+donpetry-bot@users.noreply.github.com>

don-petry requested a review from a team as a code owner June 20, 2026 00:35

don-petry mentioned this pull request Jun 20, 2026

🔁 Auto-rebase Health — Daily Report #767

Open

don-petry enabled auto-merge (squash) June 20, 2026 00:36

don-petry disabled auto-merge June 20, 2026 00:37

don-petry enabled auto-merge (squash) June 20, 2026 00:37

gemini-code-assist Bot reviewed Jun 20, 2026

View reviewed changes

Comment thread scripts/dev-lead-fix-reviews.sh

don-petry disabled auto-merge June 20, 2026 00:38

coderabbitai Bot previously approved these changes Jun 20, 2026

View reviewed changes

fix(reviews): address review comments [skip ci-relay]

cbe4442

don-petry dismissed coderabbitai[bot]’s stale review via cbe4442 June 20, 2026 00:43

don-petry enabled auto-merge (squash) June 20, 2026 00:43

don-petry disabled auto-merge June 20, 2026 00:45

don-petry enabled auto-merge (squash) June 20, 2026 00:46

donpetry-bot approved these changes Jun 20, 2026

View reviewed changes

don-petry merged commit cfbe24e into main Jun 20, 2026
38 checks passed

don-petry deleted the claude/github-discussion-735-yvc0iq branch June 20, 2026 01:37

don-petry mentioned this pull request Jun 21, 2026

ci(dev-lead): pin self-host caller to dev-lead/next (ring-0 canary) #859

Merged

github-actions Bot mentioned this pull request Jun 24, 2026

petry-projects — workflow failures detected 2026-06-24 #917

Closed

github-actions Bot mentioned this pull request Jun 2, 2026

petry-projects — workflow failures detected 2026-06-02 #419

Closed

Navigation Menu

Repository navigation

fix(dev-lead): repair auto-rebase conflict detection and applied-rate signal - #812

fix(dev-lead): repair auto-rebase conflict detection and applied-rate signal#812
don-petry merged 2 commits into
mainfrom
claude/github-discussion-735-yvc0iq

don-petry commented Jun 20, 2026

Uh oh!

chatgpt-codex-connector Bot commented Jun 20, 2026

Uh oh!

coderabbitai Bot commented Jun 20, 2026 •

edited

Loading

Review limit reached

Uh oh!

don-petry commented Jun 20, 2026

Uh oh!

donpetry-bot commented Jun 20, 2026

Uh oh!

don-petry commented Jun 20, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

Uh oh!

don-petry commented Jun 20, 2026

Uh oh!

sonarqubecloud Bot commented Jun 20, 2026

Uh oh!

don-petry commented Jun 20, 2026

Uh oh!

donpetry-bot left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

don-petry commented Jun 20, 2026

Why

Root cause

Changes

Verification

Note for the initiative (epic #736 / story #739)

Uh oh!

chatgpt-codex-connector Bot commented Jun 20, 2026

Uh oh!

coderabbitai Bot commented Jun 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review limit reached

Uh oh!

don-petry commented Jun 20, 2026

Dev-Lead — review-changes (no-changes)

Uh oh!

donpetry-bot commented Jun 20, 2026

Uh oh!

don-petry commented Jun 20, 2026

Dev-Lead — fix-bot-comment (no-changes)

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

don-petry commented Jun 20, 2026

Dev-Lead — fix-reviews (applied)

Uh oh!

sonarqubecloud Bot commented Jun 20, 2026

Quality Gate passed

Uh oh!

don-petry commented Jun 20, 2026

Dev-Lead — review-changes (no-changes)

Uh oh!

donpetry-bot left a comment

Choose a reason for hiding this comment

Automated review — APPROVED ✓

Summary

Linked issue analysis

Findings

CI status

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

coderabbitai Bot commented Jun 20, 2026 •

edited

Loading