agentsclimarketplace
Sign in

Review spec

Skill GrinRus/ai_driven_dev/skills/review-spec

Reviews plan and PRD, then gates readiness for task derivation. Use when plan and PRD artifacts are ready for downstream planning readiness. Do not use when the request is plan authoring in `plan-new` or task derivation in `tasks-new`.From its SKILL.md

Install
npx -y skills add GrinRus/ai_driven_dev --skill review-spec

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.0 KB, as published. Nobody here has run it

Follow feature-dev-aidd:aidd-core.

Steps

  1. Set active stage review-plan, then review-prd; keep active feature in sync via canonical active-state runtime commands only.
  2. Run plan readiness gate before verdict: python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-core/runtime/plan_review_gate.py --ticket <ticket>.
  3. Canonical plan path for all checks and narrative is only aidd/docs/plan/<ticket>.md; alias paths like aidd/docs/plan/<ticket>.plan.md are forbidden.
  4. Run python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-core/runtime/prd_review.py --ticket <ticket> to refresh structured PRD report payload.
  5. Gate PRD readiness with python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-flow-state/runtime/prd_check.py; block on failure.
  6. Use the existing rolling context pack as review input; do not invoke standalone context-pack builder scripts from this stage.
  7. Run subagent feature-dev-aidd:plan-reviewer. Treat its output as narrative review input only; if the plan gate already yields a terminal blocker, keep that blocker authoritative for the stage verdict.
  8. Run subagent feature-dev-aidd:prd-reviewer after the plan-reviewer handoff whenever PRD artifacts remain readable; if the plan path is WARN-only, continue the PRD pass, but do not let narrative reviewers override gate/report payloads.
  9. Persist PRD review report with python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-core/runtime/prd_review.py --ticket <ticket> --report aidd/reports/prd/<ticket>.json --require-ready for READY-path validation.
  10. Review-stage RLM policy: if RLM links are empty, the runtime performs bounded auto-heal (rlm_links_build, then finalize probe) and returns WARN attribution with evidence instead of terminal block when review can continue.
  11. Report payload (aidd/reports/prd/<ticket>.json|.pack.json) is the source of truth for final verdict: READY is allowed only when recommended_status=ready; otherwise return WARN/BLOCKED with canonical next action on upstream PRD/plan fixes, then rerun /feature-dev-aidd:review-spec <ticket>.
  12. If narrative text conflicts with gate/report payload, mark WARN(review_spec_report_mismatch) and follow gate/report payload only.
  13. Proceed to implementation is forbidden when recommended_status != ready.
  14. Runtime-path safety: execute only canonical runtime commands from this contract (python3 ${CLAUDE_PLUGIN_ROOT}/skills/.../runtime/...).
  15. Root-relative /skills/... runtime paths are forbidden; use only python3 ${CLAUDE_PLUGIN_ROOT}/skills/.../runtime/....

Command contracts

python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-core/runtime/prd_review.py

  • When to run: as canonical review-spec stage entrypoint before PRD approval decisions.
  • Inputs: --ticket <ticket> plus active PRD/plan artifacts.
  • Outputs: normalized PRD review report payload; verdict decisions use report payload (recommended_status) as source of truth.
  • Failure mode: non-zero exit when review inputs are incomplete or report validation fails; RLM links-empty path is handled as bounded auto-heal then WARN attribution.
  • Next action: fix missing artifacts/review findings or unresolved RLM prerequisites and rerun the CLI.

python3 ${CLAUDE_PLUGIN_ROOT}/skills/aidd-core/runtime/plan_review_gate.py

  • When to run: before the plan-reviewer narrative pass and before deciding whether the stage can continue on a WARN path.
  • Inputs: --ticket <ticket> and the canonical plan artifact.
  • Outputs: structured plan gate verdict used as an authoritative source for early-stop semantics.
  • Failure mode: non-zero exit when the plan artifact is missing, invalid, or blocked by the gate contract.
  • Next action: fix the plan blockers and rerun the gate before claiming review-spec readiness.

Notes

  • Planning stage: AIDD:ACTIONS_LOG: n/a.

Additional resources

Keep looking

Skills are one crate of 627,559. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.