Reject reviewer handoffs with contradictory mutation ledger or ambiguous final decision state #501

New Issue

2026-07-08T02:25:00-05:00

jcwalker3 commented

2026-07-08 02:25:00 -05:00

Problem

A reviewer handoff for PR #472 reported that the single terminal mutation budget had already been consumed by merging PR #411, but the mutation ledger only listed acquiring the PR #472 reviewer lease. The same handoff also claimed Gitea review submission was blocked, while the tool log showed gitea_mark_final_review_decision was invoked. This leaves the next controller/reviewer unable to know whether the review state was actually mutated, whether the terminal mutation budget was consumed, or whether the approval is safe to submit.

Required behavior

Reviewer/final-review handoffs must be rejected or flagged when they contain inconsistent mutation claims.
If a merge, close, approve, request-changes, final-decision mark, lease acquisition, label change, branch update, or comment post occurs, it must appear in the mutation ledger.
If a tool fails closed before mutation, the handoff must explicitly say:
- tool called
- mutation attempted
- mutation rejected
- no server-side state changed
- proof/source of that conclusion
If the model claims a single-mutation gate blocked action, the handoff must identify the exact prior mutation that consumed the budget and include it in the ledger.
Final-review state must distinguish between:
- local intent prepared
- server-side final decision marked
- Gitea review submitted
- Gitea review rejected/blocked

Acceptance criteria

Add validation for canonical reviewer handoff comments that detects contradiction between narrative claims and mutation ledger.
Add tests covering:

claimed merge missing from mutation ledger
claimed review blocked but final decision marked
lease listed but review decision omitted
attempted mutation rejected with no proof

Ensure the validator fails closed and asks the LLM to rewrite the handoff before posting.
Add a ready-to-paste corrected handoff template for blocked review submissions.
Confirm existing reviewer/merge safety gates still pass.

## Problem A reviewer handoff for PR #472 reported that the single terminal mutation budget had already been consumed by merging PR #411, but the mutation ledger only listed acquiring the PR #472 reviewer lease. The same handoff also claimed Gitea review submission was blocked, while the tool log showed `gitea_mark_final_review_decision` was invoked. This leaves the next controller/reviewer unable to know whether the review state was actually mutated, whether the terminal mutation budget was consumed, or whether the approval is safe to submit. ## Required behavior - Reviewer/final-review handoffs must be rejected or flagged when they contain inconsistent mutation claims. - If a merge, close, approve, request-changes, final-decision mark, lease acquisition, label change, branch update, or comment post occurs, it must appear in the mutation ledger. - If a tool fails closed before mutation, the handoff must explicitly say: - tool called - mutation attempted - mutation rejected - no server-side state changed - proof/source of that conclusion - If the model claims a single-mutation gate blocked action, the handoff must identify the exact prior mutation that consumed the budget and include it in the ledger. - Final-review state must distinguish between: - local intent prepared - server-side final decision marked - Gitea review submitted - Gitea review rejected/blocked ## Acceptance criteria 1. Add validation for canonical reviewer handoff comments that detects contradiction between narrative claims and mutation ledger. 2. Add tests covering: - claimed merge missing from mutation ledger - claimed review blocked but final decision marked - lease listed but review decision omitted - attempted mutation rejected with no proof 3. Ensure the validator fails closed and asks the LLM to rewrite the handoff before posting. 4. Add a ready-to-paste corrected handoff template for blocked review submissions. 5. Confirm existing reviewer/merge safety gates still pass.

jcwalker3 referenced this issue from a commit

2026-07-08 02:33:36 -05:00

feat: reject contradictory reviewer handoff ledgers (Closes #501)

jcwalker3 referenced a pull request that will close this issue

2026-07-08 02:33:44 -05:00

feat: reject contradictory reviewer handoff ledgers (Closes #501) #506

sysadmin referenced this issue

2026-07-08 22:22:39 -05:00

feat: reject contradictory reviewer handoff ledgers (Closes #501) #506

sysadmin referenced this issue

2026-07-09 07:51:38 -05:00

feat: reject contradictory reviewer handoff ledgers (Closes #501) #506