skills/review-agent/SKILL.md
Launch a sub-agent to review the current branch diff against master
npx skillsauth add api-haus/my-claude-workflow review-agentInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Launch a sub-agent to code-review the current branch. The agent gathers its own data and performs the review independently.
Use the Agent tool with subagent_type "general-purpose" and this prompt verbatim:
You are a senior code reviewer. Your task: review the current branch as a pull request.
## Step 1: Gather data
Run these commands yourself to collect what you need:
1. `git log --oneline origin/master...HEAD` — commit log
2. `git diff origin/master...HEAD -- . | cat` — full diff (pipe through cat to avoid pager)
3. Read the CLAUDE.md in the repo root if it exists (use Glob to find it)
If origin/master doesn't exist, try origin/main instead.
## Step 2: Read context
Read any files from the diff that you need more context on to understand the changes. Don't just review the diff in isolation — open the actual files to see surrounding code, types, and patterns.
## Step 3: Review
Provide a thorough code review covering:
1. **What does this change do?** — Summarize from the diff
2. **Correctness** — Logic bugs, edge cases, off-by-one errors, type safety
3. **Test quality** — Do tests prove the feature works? Could they pass with a broken implementation? Missing scenarios?
4. **Code quality** — Naming, duplication, unnecessary complexity, dead code
5. **Performance** — O(n²) loops, unnecessary allocations, missing indexes
6. **Security** — Injection, auth bypass, data exposure
7. **Convention compliance** — Does it follow project CLAUDE.md rules?
Be critical. Flag anything suspicious. Rate severity: blocking / warning / nit.
Format with clear sections and a summary table.
data-ai
Extract research content from YouTube presentations, PDFs, or PPTX files into structured markdown. Dispatches each pass to a dedicated sub-agent (research-extractor / research-vision / research-refiner) so per-deck vision passes scale to hundreds of slides without bloating the parent context.
development
Build, run, and analyze Unity profiler data with perf-report-style call-stack attribution
documentation
Write a handoff prompt for a future session. A handoff is a continuation-link — minimal context plus a kickoff line the user can copy-paste. Never a diagnosis, never an investigation script, never a prescribed deliverable.
testing
Multi-agent orchestration mode. The orchestrator never reads, edits, runs, or tests directly — it scopes work, runs a re-implementation audit, presents a freeform method brief with grounded recommendations, then dispatches every step to sub-agents through shared context files at `docs/orchestrate/<topic>/`. Use when invoked via /delegate, when the user asks to orchestrate or coordinate multi-agent work, or when the task explicitly calls for delegation.