What problem does it solve? Addressing pull request review feedback is tedious and unbounded: AI reviewers like GitHub Copilot generate endless suggestions, some valid and some out of scope, and keeping CI green while replying to every thread is manual, repetitive work. ## Core Features & Use Cases - Iterative review loop: Fetches unresolved review threads, triages each comment by scope, implements accepted fixes, commits, pushes, replies, resolves Copilot threads, and re-requests review until Copilot is satisfied. - Scope and bloat guardrails: Captures the PR's intent and original size, declines out-of-scope or stylistic suggestions with reasoning, and enforces a bloat budget so review fixes never balloon the PR. - CI gate integration: Waits for required checks on every pushed commit, diagnoses failures via gh CLI, fixes PR-caused failures, and never weakens checks to get green. - Human decision points: Flags intent-changing or unverifiable comments and asks the user before acting, rather than posting anything prematurely on GitHub. - Use Case: After opening a PR, invoke this skill to automatically work through Copilot's review rounds — implementing valid fixes, pushing back on bad suggestions, keeping CI green — until Copilot approves or has no further comments. ## Quick Start Run the github-resolve skill on my current pull request to address all Copilot and human review comments until the review is clean and CI passes.
