avibebuilder/claude-prime / .claude/skills/review-code
review-code
VerifiedReview code for quality, correctness, and fit. Use when the user wants judgment on code that already exists — their own changes, a teammate's patch, a PR, branch, commit, diff, staged changes, or one or more files to look over. Activate on requests like review, look over, sanity check, critique, code review, or 'is this good?' The key signal is that the user wants evaluation of existing code and its tradeoffs, not implementation, debugging, or explanation. This skill works independently, but when plans, specs, task artifacts, or prior discussion exist, use them to understand why the code exists before judging it.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 2.8 KB |
| teammate.md | markdown | 3.4 KB |
teammate.md
View on GitHub# Reviewer handoff (`/review-code` in an isolated teammate)
When a caller just authored the code being reviewed, spawn a **reviewer** — an isolated teammate that runs `/review-code` on the caller's behalf.
## Why isolation matters
The agent that just authored the code carries the narrative of *what it was trying to do*. That narrative biases judgment. A teammate with zero prior context evaluates the code on its merits, not on the intent behind it.
## Primary mechanism: TeamCreate
Create a team with `TeamCreate` named for the unit of work (e.g., `review-auth-refactor`).
Review is read-only — do not use `isolation: "worktree"`, a plain agent is cheaper.
## Fallback: Agent subagent
If team infrastructure is unavailable, spawn directly via `Agent` with `subagent_type: "general-purpose"` and `skills: review-code` and no worktree. You lose team persistence but keep the context isolation.
## The reviewer MUST use the `/review-code` skill
The whole point is that the teammate applies `/review-code`'s process and report format. Make this non-negotiable in the prompt:
> Read `.claude/skills/review-code/SKILL.md` first and follow it exactly — it is your working instructions. If the `/review-code` skill is available as a Skill tool in your harness, invoke it. Do not improvise.
A teammate that skips the skill defeats the purpose.
## Collaborate, don't just delegate
The teammate is a collaborator, not a one-shot worker. Encourage it to:
- Ask when the claim, scope, or target is ambiguous instead of guessing
- Flag assumptions the prompt forced it to make
- Push back on framing, payload gaps, or missing context
- Surface findings mid-task when waiting to the end would waste effort
On the team path, use `SendMessage` for back-and-forth. On the subagent path, the teammate has only its return message — it surfaces questions and pushback there. Either way, treat teammate questions as the collaboration working, not a sign that it's stuck.
## What the prompt must include
- **Review target** — explicit paths, branch, or diff scope (staged, branch-vs-main, specific commits)
- **Intent** — the *what*, not the *why*. "Adds SSO via OIDC" is fine. "I chose OIDC because SAML felt too complex" is not.
- **Acceptance criteria or plan** — if one exists, link or paste it so the reviewer can check scope fit
- **Documented tradeoffs** — only those already captured in code comments, commit messages, or a linked plan
## What to leave out
- Your design justifications — they anchor the teammate's judgment
- Your confidence in the work — the teammate forms its own
- Pre-emptive defense of anything the teammate might flag
If context genuinely matters, the code itself should carry it (comment, docstring, commit message). If it doesn't, that gap is itself a finding.
## What the teammate returns
A review report per `/review-code`'s output format: Verdict, Critical Issues, Suggestions, Questions, Summary. Don't ask for a rewrite — the skill doesn't produce corrected code.
## Team lifecycle
- **Keep the teammate alive across rounds of the same work.** Message the existing teammate with the next round's input instead of re-spawning — accumulated context makes each round sharper.
- **New unit of work → new team.** When switching to a different feature or bug, shut the old team down via `SendMessage` with `{type: "shutdown_request"}` and create a fresh one. Stale context from prior work is worse than starting clean.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |