review-plan
Installation
SKILL.md
Skill: review-plan
A review gate run after planning artifacts (prd.md / design.md / implement.md) exist but before task.py start. You dispatch a sub-agent to run a structured four-dimension planning audit, then independently judge its findings — accepting those that hold up to evidence, and rejecting false positives with a concrete reason.
This prevents two failure modes: (1) disconnections / vagueness / spec conflicts shipping into implementation undetected, and (2) over-reacting to sub-agent misreads by accepting every reported issue at face value.
Step 1: Gather context
Before dispatching, collect the exact file paths the sub-agent needs. These are the context items:
| Category | What to include |
|---|---|
| Task artifacts | prd.md, design.md, implement.md — the three documents being audited |
| Reference files | Any external design reference (e.g. ref/layout/1.md for a UI task), current source code the task modifies |
| Relevant specs | Every .trellis/spec/game/*.md that the design touches (minimum: theming.md, navigation.md, game-state.md; add domain-specific specs as needed) |
| Architecture vocabulary | Load the codebase-design skill for its leading words (module, interface, depth, seam, locality, leverage, deletion test); the sub-agent must use these terms when reporting architecture depth findings (Dimension D) |
| Process docs | .trellis/workflow.md — for task-planning quality standards |