Skill v1.0.1
currentLLM-judged scan95/100+5 new
version: "1.0.1" name: review description: Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X". Reviews changes along two axes — Code Review (does the code follow this repo's documented coding standards?) and Spec Review (does the code match the originating issue/PRD?). Runs both axes as parallel sub-agents and reports them side by side.
Review
Overview
Two-axis review of the diff between HEAD and a fixed point. Both axes run as parallel sub-agents so they don't pollute each other's context, then this skill aggregates their findings.
Core Workflow
Step 1 — Determine args
Goal: Resolve named params into review-diff args and spec path.
Exit criterion: ref: and spec: resolved; ready to pass to sub-agents.
| Param | Meaning | |
|---|---|---|
ref:<ref> | Diff base; passed to review-diff | |
spec:<path> | Path to the spec file (issue or PRD); passed to the Spec Review agent |
`ref:` values:
| Value | review-diff args | |
|---|---|---|
ref:dirty | All dirty files in repo | |
ref:worktree | All changed files since start of worktree | |
ref:<branch> | All changed files since start of <branch> | |
ref:<commit> | All changed files in <commit> | |
ref:<start>..<end> | All changed files between <start> and <end> |
If spec: or ref: is absent and cannot be inferred from context, ask the user — do not silently infer.
Step 2 — Spawn sub-agents
Goal: Run Code Review and Spec Review in parallel without polluting each other's context.
Exit criterion: Both sub-agents have returned their reports.
Send a single message with two Agent tool calls. Use the general-purpose subagent for both.
- Code Review agent: read
references/code-agent.mdfor the full brief; - Spec Review agent: read
references/spec-agent.mdfor the full brief;
Step 3 — Aggregate
Goal: Present both reports side by side without merging or reranking.
Exit criterion: Both reports displayed; one-line summary given.
Present the two reports under ## Code Review and ## Spec Review headings, verbatim or lightly cleaned. Do not merge or rerank findings — the two axes are deliberately separate.
End with a one-line summary: total findings per axis, and the worst single issue (if any) flagged.
Step 4 — Handback
Goal: Return control to the calling context.
Exit criterion: Caller workflow resumed, or session ended.
- If this skill was invoked by another skill (e.g.
/ralph): - Resume the calling skill's workflow immediately
- Do not stop, do not wait for user input, do not end your turn.
- If this review was the only task requested:
- Stop here and wait for user input.
Common Rationalizations
| Rationalization | Reality | |
|---|---|---|
| "The two findings overlap, I'll merge them into one report" | Merged reports hide which axis a finding belongs to. Keep them separate. | |
| "Spec findings are minor, not worth reporting" | Spec drift compounds. Report everything; let the user dismiss. |
Checklist
- [ ] Args determined from user intent
- [ ] Both sub-agents spawned in a single parallel message
- [ ] Each sub-agent ran
review-diffitself — no diff pasted inline - [ ] Reports presented under separate
## Code Reviewand## Spec Reviewheadings - [ ] One-line summary at the end