bug: auto-merge never triggers for bot-authored PRs — selfApprovalFallback downgrades APPROVE→COMMENT but publishCriteriaApprovedReview requires APPROVE

[baderdean]

Summary

When looper runs as a bot account (e.g. ankaboot-bot) and the worker opens PRs as that bot, the reviewer cannot APPROVE its own PRs (GitHub blocks self-approval). Looper correctly downgrades APPROVE → COMMENT via selfApprovalFallback, but the auto-merge path requires marker.Event == ReviewEventApprove and returns early on COMMENT — so EnableAutoMerge is never called.

Evidence

PR: ankaboot-source/m3llm#180 (bot-authored, reviewer-clean)

• Reviewer loop Add a Looper inbox for completed runs and needs-human items #49 completed with outcome=clean (all fixer corrections verified).
• Config: [roles.reviewer.autoMerge] enabled = true, requireBranchProtection = false, scope = "looper-only", strategy = "squash".
• reviewEvents.clean = "APPROVE".
• GitHub autoMergeRequest: null on the PR — auto-merge was never enabled.

Root cause

internal/reviewer/runner.go:

• Line 3791-3801 (submitOrReuseReview): selfApprovalFallback := event == ReviewEventApprove && sameReviewAuthorLogin(detail.Author, currentLogin) — when PR author == current user (bot), downgrades APPROVE → COMMENT. This is correct (GitHub forbids self-approval).

• Line 3672 (publishCriteriaApprovedReview): if marker.Event != ReviewEventApprove { return ... } — since marker.Event is COMMENT (downgraded), returns early and never calls decideAutoMerge or EnableAutoMerge.

The selfApprovalFallback correctly handles the review event, but the auto-merge path does not account for it.

Relationship to #352

#352 (auto-merge PRD) assumes Reviewer's APPROVE as the trigger. It does not address the case where the reviewer and PR author share the same GitHub identity (single-bot-account mode), where APPROVE is downgraded to COMMENT by selfApprovalFallback.

Relationship to #598

Same root cause as #598 (single-identity mode): the bot can review (COMMENT) but cannot approve or auto-merge its own PRs.

Expected behavior

When autoMerge.enabled = true and the reviewer marks the PR outcome=clean, auto-merge should be enabled via GitHub's native auto-merge API (gh pr merge --auto), even if the review event was downgraded to COMMENT by selfApprovalFallback. GitHub's branch protection (not the review event type) is the authority for "safe to merge" per #352.

Proposed fix

In publishCriteriaApprovedReview (runner.go:3672), relax the gate:

// Before
if marker.Event != ReviewEventApprove { return ... }

// After
if marker.Event != ReviewEventApprove && !selfApprovalFallback { return ... }

Or, more robustly: gate on outcome=clean (the reviewer's verdict) rather than marker.Event (the GitHub review event type), since selfApprovalFallback already correctly downgrades the event type for GitHub API compliance.

Environment

• looper 0.0.0-dev (dev build, EBADF workaround for review submit fails with EBADF: auto-upgrade hook consumes the single-use trusted review config FD #595)
• opencode 1.17.20
• GitHub, daemon running as bot account (ankaboot-bot)
• Config: roles.reviewer.autoMerge.enabled = true, requireBranchProtection = false

Reproduction

1. Run looperd as a bot account (GH_TOKEN=bot_pat in daemon env)
2. Worker opens a PR as the bot
3. Reviewer reviews and marks outcome=clean
4. Review event is downgraded to COMMENT (selfApprovalFallback)
5. publishCriteriaApprovedReview returns early (marker.Event != APPROVE)
6. EnableAutoMerge never called → PR sits at human gate indefinitely

Context

• PRD: Reviewer auto-merge for Looper PRs via GitHub-native auto-merge #352 (auto-merge PRD)
• feat: single-user / self-review mode — reviewer discovery for worker PRs without a separate bot account #598 (single-user / self-review mode)
• review submit fails with EBADF: auto-upgrade hook consumes the single-use trusted review config FD #595 (EBADF, fixed by dev build)

[alucero270]

This affects my single-identity setup — self-review PRs can never auto-merge. Happy to pick this up and submit a fix this week.