github-pr-review
Evaluate GitHub CodeQL and Copilot suggestions on PRs — fetch, triage, propose fixes.
- License 许可
- MIT
- 最近更新
- 2026-11-13
- Stars
- ★ 2
- Forks
- 2
01 前言 Front Matter
- name
- github-pr-review
- author
- Hermes Agent
- license
- MIT
- version
- 1.0.0
- metadata
- { "hermes": { "tags": [ "GitHub", "PR", "CodeQL", "Copilot", "Review", "Security", "Code Quality" ], "related_skills": [ "github-codespace", "github-code-review" ] } }
- platforms
- [ "linux" ]
- description
- Evaluate GitHub CodeQL and Copilot suggestions on PRs — fetch, triage, propose fixes.
03 SKILL.md 正文
GitHub PR Review: CodeQL & Copilot Suggestions
Evaluates automated review comments from GitHub Advanced Security (CodeQL) and GitHub Copilot on pull requests. Fetches comments, triages by severity, and proposes fixes with a decision framework.
Prerequisites
- GitHub API access (token via
/proc/PID/environorgh auth) - PR number or URL
Workflow
Step 1: Fetch Review Comments
# Get PR review comments (CodeQL + Copilot inline comments)
curl -s "https://api.github.com/repos/{owner}/{repo}/pulls/{PR}/comments" \
-H "Authorization: token $TOKEN" | python3 -c "
import json, sys
comments = json.load(sys.stdin)
for c in comments:
author = c.get('user', {}).get('login', '?')
if 'advanced-security' in author or 'copilot' in author:
print(f'---')
print(f'File: {c.get(\"path\", \"?\")}')
print(f'Line: {c.get(\"line\", \"?\")}')
print(f'Author: {author}')
print(f'Body: {c.get(\"body\", \"\")[:500]}')
"
Step 2: Triage by Source
| Source | What It Checks | Typical Findings |
|---|---|---|
github-advanced-security[bot] |
CodeQL static analysis | Security vulnerabilities, workflow permissions, injection risks |
Copilot |
Code quality suggestions | Inconsistent naming, dead code, style issues |
copilot-pull-request-reviewer |
PR-level review | Logic errors, missing tests, documentation |
Step 3: Evaluate Each Suggestion
For each comment, apply this decision framework:
ACCEPT if:
- Security issue (CodeQL finding) → always accept or explicitly justify rejection
- Workflow permission too broad → add
permissions:block - Actual bug or logic error → fix the code
- Inconsistent naming/branding → fix (e.g., wrong logo on badge)
REJECT if:
- False positive (CodeQL misidentification) → document why
- Style preference only (no functional impact) → note but skip
- Would break existing behavior without clear benefit → reject
- Suggestion is overly aggressive (e.g., "rename all variables") → reject
DEFER if:
- Requires architectural change → propose in separate PR
- Needs user decision → present options, wait for approval
- Unclear impact → investigate further before acting
Step 4: Present Proposal
Format as a table for user review:
=== CodeQL & Copilot Review: PR #N ===
| # | Severity | Source | File | Suggestion | Proposal | Effort |
|---|----------|--------|------|------------|----------|--------|
| 1 | High | CodeQL | workflow.yml | Missing permissions block | ACCEPT — add permissions: contents: read | 2 lines |
| 2 | Low | Copilot | README.md | Wrong badge logo | ACCEPT — fix logo parameter | 1 line |
| 3 | Medium | CodeQL | script.sh | Unquoted variable | REJECT — variable is intentionally word-split | 0 lines |
Summary: 2 accept, 1 reject, 0 defer
Step 5: Implement Accepted Fixes
After user approval:
- Make the changes
- Commit with descriptive message
- Push and verify CI passes
- Optionally reply to the review comment with the fix
Step 6: Seed Export Check (Pre-Merge)
Before merging, check if new high-value Mnemon entries should be added to seed.json:
scripts/check-seed-export.sh
This script:
- Queries Mnemon for entries with importance >= 4
- Compares against existing seed.json content
- Reports any new entries that should be exported
If new entries found:
- Review each entry for relevance
- Add to
mnemon/seed.jsonwith proper format - Validate:
mnemon import --dry-run mnemon/seed.json - Commit updated seed.json with the PR
Why this matters:
- Seed file is the persistent knowledge catalog across sessions
- New wiki articles, key decisions, and architecture facts should be seeded
- Ensures future Codespaces start with all important knowledge
API Endpoints
| Endpoint | Purpose |
|---|---|
GET /repos/{owner}/{repo}/pulls/{PR}/comments |
Inline review comments |
GET /repos/{owner}/{repo}/issues/{PR}/comments |
Top-level PR comments |
GET /repos/{owner}/{repo}/pulls/{PR}/reviews |
Full review objects |
POST /repos/{owner}/{repo}/pulls/{PR}/comments/{ID}/replies |
Reply to a comment |
Common CodeQL Findings
Workflow Permissions
# BEFORE (triggers CodeQL warning)
name: CI
on: [push]
jobs:
build:
runs-on: ubuntu-latest
steps: [...]
# AFTER (fixes CodeQL warning)
name: CI
on: [push]
permissions:
contents: read
jobs:
build:
runs-on: ubuntu-latest
steps: [...]
Script Injection
# BEFORE (vulnerable)
- run: echo "${{ github.event.issue.body }}"
# AFTER (safe)
- run: echo "$ISSUE_BODY"
env:
ISSUE_BODY: ${{ github.event.issue.body }}
Hardcoded Secrets
# BEFORE (exposed)
env:
API_KEY: sk-abc123...
# AFTER (use secrets)
env:
API_KEY: ${{ secrets.API_KEY }}
Common Copilot Findings
- Inconsistent naming: Badge uses wrong logo, variable name mismatches
- Dead code: Unused imports, unreachable branches
- Missing error handling: Unchecked return values
- Style: Inconsistent formatting, missing documentation
Pitfall — Copilot quota gate: The Copilot reviewer may decline to review with
"insufficient remaining quota" — this is an account-level limit, not a PR-level
problem. Do not wait on it or retry; treat it as "no Copilot review available"
and proceed with CodeQL + manual review. The review comment still appears in
the reviews list as COMMENTED with the quota message, so filter by author
and body before treating a review as substantive.
2. Don't fix style issues in functional code changes — keep PRs focused
3. Don't ignore security findings — even if you reject, document why
4. Don't reply to bot comments unless meaningful — bots don't read replies
5. Check for duplicate comments — CodeQL often comments on same issue twice
Quick Reference
# Fetch all review comments for a PR
curl -s "https://api.github.com/repos/{owner}/{repo}/pulls/{PR}/comments" \
-H "Authorization: token $TOKEN" | jq '.[] | {path, line, user: .user.login, body: .body[0:200]}'
# Count comments by source
curl -s "https://api.github.com/repos/{owner}/{repo}/pulls/{PR}/comments" \
-H "Authorization: token $TOKEN" | jq -r '.[].user.login' | sort | uniq -c
# Reply to a comment
curl -s -X POST "https://api.github.com/repos/{owner}/{repo}/pulls/{PR}/comments/{COMMENT_ID}/replies" \
-H "Authorization: token $TOKEN" \
-d '{"body": "Fixed in commit abc123"}'
04 同仓库更多 Skills
version-single-source
Use when adding a new component dependency or versioning a binary in install/boot scripts.
tdd
Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.
shell-path-login
Use when a binary or script works in one shell context but not another (interactive vs non-interactive login vs non-login shell).
self-update-fix
Fix a failing dependency update PR. Use when a self-update PR's CI fails, or when invoked by scripts/self-update-fix.sh.
pr-change-log
Use when a multi-commit PR needs its description written/updated, or when closing out a long branch with many fix commits.
persistent-knowledge
Persistent skills/knowledge in Codespace via symlinks.