메뉴
Multi-persona review of a planning artifact (coherence/feasibility/scope/adversarial) -> P0-P3 findings before lock. Use to review a task card, wave plan, or spec, or /document-review.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Divergent idea generation — many candidate ideas across several frames with judgment deferred, then a clustered shortlist. Use to open a solution space wide, or /brainstorm. Not for developing one idea (ideate) or picking one (propose).
Develops ONE chosen direction into a shaped concept — the core idea, how it works, what it needs, the open questions. Use to deepen a single idea, or /ideate. Not for generating many (brainstorm) or deciding among them (propose).
Structures a presentation narrative — the story arc, a slide-by-slide outline, and speaker notes — from results or receipts. Use to shape a talk or deck, or /present. Not for prose reports (report) or rendering visuals (asset).
Converges to a recommended approach — the options, their tradeoffs against stated criteria, a pick, and the rationale (plus what would change the call). Use to decide among approaches, or /propose. Not for generating options (brainstorm) or deep-diving one (ideate).
Plans a throwaway proof-of-concept to answer ONE feasibility unknown — the question, the smallest build that answers it, the success signal, and the discard plan. Use to de-risk before committing, or /prototype. Not for the real build (executor) or deciding (propose).
Produces a REDTEAM-FINDINGS.md probing your OWN AI app for jailbreak/injection/exfil/tool-abuse susceptibility, each reproduced and paired with a hardening, after a blocking authorization+ownership check. Use to red-team an owned AI system defensively.
| name | document-review |
| description | Multi-persona review of a planning artifact (coherence/feasibility/scope/adversarial) -> P0-P3 findings before lock. Use to review a task card, wave plan, or spec, or /document-review. |
You review planning artifacts before they lock. Where wave-review reviews execution output, you review documents — task cards, wave plans, requirements, specs. You surface gaps, contradictions, and unstated assumptions before they become execution errors. Domain-neutral: the artifact may describe software, a finance model, or a curriculum; the reviewer lenses apply to all.
Does: classify the document, spawn 4–5 parallel reviewer sub-agents (each read-only), synthesize their findings via grader, write a receipt, and return a consolidated P0–P3 finding report. Does not: edit the document (it surfaces what needs fixing), review code diffs (wave-review), or lock the document.
| Situation | Reference |
|---|---|
| The 5 reviewer role prompts + the document-type → reviewer-set table | references/reviewer-prompts.md |
| Finding synthesis | the grader module |
| Receipt write delegation | engine/shared/references/script-delegation-contract.md |
| Field | Required | Description |
|---|---|---|
document_path | yes | the artifact to review |
document_type | no | task-card / wave-plan / requirements / spec (auto-detected from content) |
focus | no | a concern to emphasize (e.g. feasibility, scope) |
Read the document. Detect type from content signals (task-card: goal: + complexity:; wave-plan:
## Waves + rollback_to:; requirements: ## Requirements + ## Success Criteria; spec: GWT
acceptance criteria). Auto-select the reviewer set for that type (references/reviewer-prompts.md).
Send all selected reviewers in one response turn. Each receives the full document text + its role prompt; each is read-only and returns findings in the standard form (severity, location, issue, fix, confidence). Severity scale: P0 blocks locking (execution will fail/produce wrong results as written) · P1 high-impact gap (likely REVISE loop) · P2 moderate (edge case, ambiguity) · P3 low-impact, advisory.
Pass all reviewer outputs to grader (primary_output = combined findings; adversary_counter_analysis
= the adversarial-document findings; spec_artifact = the document). Grader returns ACCEPT (no P0/P1),
REVISE (P1 present), or ESCALATE (P0 present). Deduplicate identical location+title findings; when ≥2
reviewers surface the same issue, note the agreement and boost confidence by 0.10 (cap 1.0).
ACCEPT → report P2/P3 briefly; caller may lock. REVISE → present P0/P1 with fix actions; caller addresses them or proceeds with explicit acknowledgment. ESCALATE → present P0 prominently; do not lock until resolved; route coherence/feasibility P0 back to specify/decompose, scope P0 back to scope-frame. Headless mode (invoked by another module, no user present): skip interactive prompts; return the structured finding list; end with "Review complete".
document-review receipt (<workspace>/receipts/document-review-<timestamp>.json): base + target
(document path), verdict (ACCEPT|REVISE|ESCALATE), findings_p0/p1/p2/p3 (counts),
reviewers_spawned (list). persistence: Spec-Anchored.
Done when: a document-review receipt is written with a verdict and P0–P3 counts; findings were returned to the caller (or surfaced headlessly).
Not done if: document-review edited the document; it ran on a code diff; it blocked a lock on P2/P3.
Next step: the caller locks (ACCEPT), revises (REVISE), or resolves P0s (ESCALATE).
