| name | implementation-guide-review-loop |
| description | Run a state-driven review/fix loop for implementation-guide documents only (not code), using `.cursor/agents/implementation-guide-state.md` and `.cursor/agents/implementation-guide-review-state.md`. Use when the user asks to review an implementation guide, run blocker-only guide review, or loop reviewer/fixer until guide blockers are resolved. |
Implementation Guide Review Loop
Use this skill for implementation-guide document quality loops, not code implementation loops.
Durable files:
.cursor/agents/implementation-guide-state.md - guide path and review scope.
.cursor/agents/implementation-guide-review-state.md - current-cycle blocker ledger.
New-Chat Simulation Rule
Each cycle must simulate:
new chat -> guide review -> blocker findings -> guide fix -> new chat -> guide review
Rules:
- Spawn a fresh
@implementation-guide-reviewer every review cycle.
- Do not pass prior reviewer/fixer chat content to the next reviewer.
- Fixer must reset guide-review-state to clean
READY_FOR_REVIEW after applying guide fixes.
Modes
full_guide:
- Review the entire guide from guide-state.
section:
- Review only the section selected in guide-state.
Loop behavior
- Read
.cursor/agents/implementation-guide-state.md.
- Initialize/reset
.cursor/agents/implementation-guide-review-state.md with clean active_findings: [].
- Call fresh
@implementation-guide-reviewer.
- If state ->
REVIEW_BLOCKED, call @implementation-guide-fixer.
- Fixer edits the guide and resets state to
READY_FOR_REVIEW.
- Call a new
@implementation-guide-reviewer.
- Repeat until:
COMPLETE, or
MAX_ROUNDS_REACHED (hard cap 20), or
NEEDS_CLARIFICATION.
Hard rules
- Blocker-only findings; ignore enhancements.
- Fixer edits guide docs only (plus guide-review-state reset).
- Do not modify application code in this loop.
- Keep guide structure stable unless blocker resolution requires structural change.