code-review

SkillAI & models

Code review the current branch against origin/master using 7 parallel review agents with confidence scoring.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the code-review skill

What this skill tells your AI

The instructions your AI receives, as published by the-agency-ai/the-agency in .claude/skills/code-review/SKILL.md and read by ahel’s review.

Code Review

Code review the current branch against origin/master using 7 parallel review agents with confidence scoring. Runs locally against git diff; no GitHub PR required.

Arguments

  • $ARGUMENTS: Optional branch name, PR number, or empty for current branch.

Steps

Step 1: Determine branch and diff

  1. If PR number provided: gh pr view {number} --json headRefName to get branch
  2. If branch name provided: use it directly
  3. If empty: use current branch via git rev-parse --abbrev-ref HEAD
  4. Compute diff: git diff origin/master...{branch}

Step 2: Gather CLAUDE.md files

Read all CLAUDE.md files in the repo to understand project conventions.

Step 3: Summarize changes

Run git diff --stat origin/master...{branch} and categorize changed files by area.

Step 4: Launch 7 parallel review agents

Launch all 7 as Sonnet agents in a single parallel call:

  1. CLAUDE.md compliance — Check changes against project conventions
  2. Bug scan — Logic errors, null handling, type mismatches, runtime crashes
  3. Git historygit log context, are changes consistent with commit messages?
  4. Prior review context — Check for existing review files, are prior findings addressed?
  5. Code comments — Misleading comments, missing docs for public APIs
  6. Test coverage — Coverage gaps, missing edge cases, stale assertions
  7. Test consistency — Do tests match implementation? Stale mocks?

Step 5: Score findings

Send all findings to a reviewer-scorer agent (model: haiku) for confidence scoring (0-100).

Step 6: Filter

Remove findings with score < 80. Deduplicate.

Step 7: Save review

Derive the project name from the branch (last segment after /, or the branch name itself).

Save to agency/workstreams/{workstream}/code-reviews/{workstream}-review-{YYYYMMDD-HHMM}.md where {workstream} is derived from the branch name (segment before /).

Step 8: Post to GitHub (if PR exists)

If a PR exists for the branch, offer to post comments via gh pr review.

Step 9: Present summary

Show the review results: total findings, breakdown by category, high-confidence issues.

Signals

GitHub stars
55
Forks
8
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
code-review-the-agency-ai
Source
github.com/the-agency-ai/the-agency