From copilot-review
Run the local Copilot CLI to review code, fix findings, and re-review until clean. Use when user says "local copilot review", "copilot cli review", "ask copilot to review locally", "local review loop", or "run copilot on this code". This is for the npm-installed Copilot CLI — NOT the online GitHub Copilot PR reviewer.
How this skill is triggered — by the user, by Claude, or both
Slash command
/copilot-review:local-copilot-review-loopThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Iterative loop: run ACP-based review → fix findings → re-run → repeat until clean.
Iterative loop: run ACP-based review → fix findings → re-run → repeat until clean.
Use the ACP companion script for structured, session-based reviews. Replace <SKILL_BASE_DIR> with the "Base directory for this skill" path shown above when this skill is loaded:
node "<SKILL_BASE_DIR>/scripts/copilot-acp-companion.mjs" review --base <base_ref>
Options:
--base <ref> — Git base ref for diff (e.g., main, origin/main). Omit for working-tree changes.--cwd <path> — Working directory (default: current directory)--json — Output structured JSON: { review, stopReason, base, exitCode }--timeout <ms> — Max wall-clock timeout in ms (default: 1800000 = 30 min)--idle-timeout <ms> — Cancel if no activity for this long, in ms (default: 120000 = 2 min)Examples:
# Review changes against main branch
node "<SKILL_BASE_DIR>/scripts/copilot-acp-companion.mjs" review --base main
# Review staged and unstaged changes vs HEAD (untracked files are not included)
node "<SKILL_BASE_DIR>/scripts/copilot-acp-companion.mjs" review
# JSON output with focus
node "<SKILL_BASE_DIR>/scripts/copilot-acp-companion.mjs" review --base main --json "focus on error handling"
COPILOT_ACP_ALLOW_ALL_TOOLS=1 to pass --allow-all-tools to Copilot CLI, broadening tool-execution permissions. Only enable in trusted, controlled environments.[file:line] severity (high/medium/low): descriptionIMPORTANT: This is a MANDATORY loop. You MUST keep iterating until a stop condition is met. Do NOT stop after a single iteration just because you fixed or dismissed some findings.
--base ref. For PRs use the PR base branch. For local work use main or origin/main. Note: --base performs a commit-to-commit diff (base...HEAD), so staged/unstaged working-tree changes are NOT included — commit first, or omit --base to diff against HEAD including working-tree changes.--base flag[file:line] findings (only praise, "looks good", "No issues found.", or "No changes found to review.")Two findings are the same if they reference the same file, similar line range (±10 lines), and describe the same core issue. Cosmetic wording differences don't matter. If the finding is about code you just changed, it's NOT a repeat — re-assess it.
After the loop ends, report:
Local Copilot CLI review complete.
- Iterations: N
- Findings fixed: X
- Findings dismissed (invalid): Y
- Remaining: Z (if any)
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub superyyrrzz/copilot-review-plugin-cc --plugin copilot-review