What problem does it solve?
Automated code review agents need a consistent, conservative bar for settling unresolved pull request review threads without human supervision. This Skill defines the resolution criteria for deciding, per thread, whether an ask is worth implementing and safe to implement unattended, preventing sloppy or oversized autonomous fixes.
Core Features & Use Cases
- Four Outcome Decisions: Classifies each unresolved review thread as fixed, wont_fix, already_fixed/obsolete, or escalate based on worth and safety checks.
- Worth and Safety Gates: Verifies the problem still exists at the current head, requires fixes to be provable in-session via lint and existing tests, and escalates anything needing infrastructure changes or ambiguous interpretation.
- Human Override Handling: Honors standing verdicts like SAFE TO FIX and E2E REQUIRED written by humans in the thread.
- Use Case: An unattended review-resolution agent processes a PR with 12 unresolved threads, implements three contained typo and null-check fixes, declines four speculative overengineering asks with reasons, and escalates two threads requiring schema changes to a human.
Quick Start
Apply the review-hog resolution criteria to settle each unresolved review thread on this pull request, fixing what is provably safe and escalating the rest.