deliverable-review
Skill0 starsChanged 7 months agoMITView on GitHub
Use when finishing a document deliverable (law memo, SOP, tech spec, proposal, wiki page, compliance report) before declaring it done, before wrap, or when user says "/deliverable-review", "산출물 검증", "리뷰 돌려", "검증 게이트". Dispatches to a domain-specific reviewer agent and runs a verification gate sourced from feedback memory.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- EnsoDynamics/enso-methodtheir own version, first seen Oct 11, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- EnsoDynamics/enso-method★ 7same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository13 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- check-harness-context-firstSkill
- check-harnessSkill
- councilSkill
- docxSkill
- gws-setupSkill
- medical-device-ra-qa-frameSkill
- pdfSkill
- pdf-to-mdSkill
- repo-cleanupSkill
- skill-creatorSkill
- tool-setupSkill
- wrapSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 7 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Every run should reduce the probability of a new feedback_* entry being written for the same class of mistake.
- Distinctive: line 26
Do not trigger on:
- Distinctive: line 42
Use the Read tool on the target path.
- Distinctive: line 53
meddev-compliance: if any of:
- Distinctive: line 59
spec: if any of:
- Very specific: line 90
When uncertain between two, pick the more specific one and mark confidence low.
- Distinctive: line 119
Do not pre-filter feedback memory here — the reviewer does that dynamically.
- Very specific: line 159
If the reviewer returned no FAIL items, still surface WARN items prominently — they're the ones that graduate into feedback_*.md next …
- Very specific: line 173
Name the feedback source: Every WARN or FAIL must cite which feedback_*.md or which static rule it came from.
- Very specific: line 177
Flag new failure patterns: If the reviewer notices a failure mode that isn't yet in feedback memory, end the report with ### Candidate …
- Very specific: line 209
If the named reviewer agent is not yet registered in ~/.claude/agents/:
- Distinctive: line 214
If the file path doesn't exist or is empty:
- Distinctive: line 217
If feedback memory directory is missing:
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Deliverable Review
Domain verification gate for document deliverables. The goal is to catch the failure modes that keep showing up in feedback memory — missed exception clauses, scope creep, unverified assumptions, misattributed specs — before 사용자 sees the draft, not after.
This skill is a dispatcher. The actual review happens in a reviewer agent. Your job here is: detect document type, pick the right reviewer, hand it the file + context, and format the result.
Why This Exists
사용자's 6-axis harness diagnosis put 검증 (verification) at 5/10 — the thinnest axis. The pattern: doctrine is strong ("Verify Before Done", "Problem Diagnosis First"), but there's no automation for document deliverables, so the same verification failures keep accumulating as feedback_*.md entries. This skill closes that loop by turning those feedbacks into an active pre-flight check.
Every run should reduce the probability of a new feedback_* entry being written for the same class of mistake.
When To Trigger
Invoke this skill when any of the following is true:
- 사용자 explicitly says
/deliverable-review, "산출물 검증", "리뷰 돌려", "검증 게이트", "검토 돌려" - A deliverable file under
drafts/,deliverables/,compliance/,wiki/pages/was just created or substantially edited, and 사용자 is about to conclude the session - The
wrapskill is running and detects one or more deliverable candidates in the session diff - 사용자 is about to send a document to a client, regulator, or stakeholder
Do not trigger on:
- Code files (
.py,.ts,.js, …) — those belong tosuperpowers:requesting-code-review - Scratch notes, TODO lists, session logs, secall-vault files
- Commits of reference material into
refs/orwiki/sources/
Inputs
The skill accepts one of:
- File path — absolute or relative to repo root. Primary mode.
- Multiple paths — batch mode, one reviewer call per file.
- No argument — auto-detect deliverable candidates from
git status+ recent Write/Edit history, then ask 사용자 which to review.
Dispatch Logic
Step 1 — Read the file
Use the Read tool on the target path. Grab:
- Full content (or head 400 lines + tail 100 for very long files)
- Frontmatter if present
- File path and parent directory
Step 2 — Detect type (dispatcher) and subtype
Type detection has two axes:
Axis 1 — Dispatcher type (which reviewer agent gets called). Apply heuristics in order. First match wins.
meddev-compliance if any of:
- Path contains
compliance/,refs/regulations/,refs/info-security/ - Content mentions 3+ of: 의료기기법, 약사법, 시행령, 시행규칙, 고시, 식약처, MFDS, IEC 62304, ISO 13485, ISO 14971, GMP, QMS, GCP, 약관, 계약서, SOP, TPQQ, audit, HIRA 수가
- Frontmatter
type: regulation | compliance | sop | legal-memo - File name contains
sop-,법규-,compliance-,audit-,contract-,계약-
spec if any of:
- Path contains
drafts/harness-engineering/,docs/superpowers/specs/,docs/superpowers/plans/ - Content mentions 3+ of: API, endpoint, architecture, requirement, user story, acceptance criteria, 아키텍처, 요구사항, 사양, 제안서, RFP, deliverable, milestone, 성과물
- Frontmatter
type: spec | plan | proposal - File name contains
spec-,plan-,proposal-,제안-,사양-
general (fallback) for everything else — wiki entity/concept/synthesis pages, LinkedIn drafts, notes, reports that don't fit the two above.
Axis 2 — Document subtype (which checklist rows apply inside the reviewer). Subtype is passed to the reviewer regardless of dispatcher; the reviewer uses it to gate type-specific checklist items.
Detect subtype in this order. First match wins.
| Subtype | Signals |
entity | Frontmatter type: entity; path wiki/pages/ + wiki entity shape (헤딩: 인물/회사/발언/사건); filename starts with a proper noun |
concept | Frontmatter type: concept; wiki/pages/ with concept shape (definition · examples · cross-refs) |
synthesis | Frontmatter type: synthesis; content explicitly crosses 2+ sources and derives a new claim |
comparison | Frontmatter type: comparison; two-column or side-by-side structure comparing frameworks/products/approaches |
SOP | Filename sop-, path compliance/*/SOP*, content "표준작업절차" / "Standard Operating Procedure"; frontmatter type: sop |
legal-memo | Filename legal-, 법규-, 법률검토-; path compliance/; content "법률 검토", "Legal Opinion", "법적 쟁점" |
regulation-analysis | Frontmatter type: regulation-analysis; 규정·고시 해석 중심, 조문별 적용 분석 구조. legal-memo와 구분: legal-memo는 사안에 대한 법적 의견, regulation-analysis는 규정 자체의 해석·체계 정리 |
audit-response | Filename audit-, 감사-, tpqq-; content가 질의-응답 쌍 구조, audit 지적 사항 대응 |
contract | Filename contract-, 계약-, 약관-; content "제1조", "갑은 을에게", 서명란 |
proposal | Filename proposal-, 제안-; path deliverables/*/proposal*; frontmatter type: proposal |
plan | Frontmatter type: plan; docs/superpowers/plans/; 단계적 실행 순서·마일스톤·체크포인트 구조. spec과 구분: spec은 "무엇을 만드는가", plan은 "어떤 순서로 만드는가" |
spec | Frontmatter type: spec; content에 acceptance criteria·요구사항·user story 구조 |
architecture | Filename arch-, architecture-; 구성도·다이어그램이 핵심이고 본문이 그 해설인 구조. spec과 구분: 요구사항보다 구조 자체가 주제 |
report | Filename report-, 보고서-, 현황-; 기간 요약·상태 보고 성격 |
article | drafts/linkedin/, LinkedIn 후보, op-ed·에세이 형식 |
general | None of the above match — fallback |
When uncertain between two, pick the more specific one and mark confidence low.
Subtype × Source modifier: if subtype is entity or concept AND the primary source is a 강연 / 인터뷰 / 팟캐스트 / 영상 STT, also tag source:transcript. This unlocks entity-specific rules around timestamps, STT editing markers, and speaker-time anchoring. Signals for source:transcript: frontmatter sources entry has articles/ + audio-derived filename (e.g. YYYYMMDD-*-강연.md, *-interview.md, *-transcript.md), or content contains (소스 L{line}~{line}) / (hh:mm) markers, or source file itself is a transcript.
Step 3 — Dispatch to reviewer agent
Based on dispatcher type, call the matching agent via the Agent tool:
| Dispatcher type | Agent | subagent_type |
| meddev-compliance | deliverable-meddev-compliance-reviewer | (fall back to general-purpose only if the named agent isn't registered yet) |
| spec | deliverable-spec-reviewer | same |
| general | deliverable-general-reviewer | same |
Hand the agent a structured dispatch context:
DELIVERABLE REVIEW DISPATCH
- path: {{absolute path}}
- dispatcher_type: meddev-compliance | spec | general
- subtype: entity | concept | synthesis | comparison | SOP | legal-memo | regulation-analysis | audit-response | contract | proposal | plan | spec | architecture | report | article | general
- source_modifiers: [source:transcript] (if any)
- confidence: high | medium | low
- trigger: manual | wrap | explicit
- related_docs: [optional paths to RFP, contract, prior spec, source transcript, ...]The reviewer will gate its static checklist on subtype — only rows whose Type column matches the detected subtype (or all) are active. source:transcript unlocks additional transcript-specific rows for entity/concept.
Do not pre-filter feedback memory here — the reviewer does that dynamically. The dispatcher only passes the structured context.
Step 4 — Format the result
The reviewer returns a structured report. Present it to 사용자 as:
## Deliverable Review — {{filename}}
Reviewer: {{agent_name}}
Dispatcher type: {{meddev-compliance | spec | general}}
Subtype: {{entity/SOP/proposal/...}} ({{high/medium/low}})
Source modifiers: {{none | source:transcript}}
Trigger: {{manual/wrap/explicit}}
Feedback memory loaded: {{N}} entries
Checklist activated / skipped: {{k}} / {{m}} (type-gated)
### PASS ({{n}})
- ✅ {{item}}
### WARN ({{n}})
- ⚠️ {{item}} — {{reason}} [{{evidence_path}}:{{line}}]
Related feedback: {{feedback_file}}
### FAIL ({{n}})
- ❌ {{item}} — {{reason}} [{{evidence_path}}:{{line}}]
Related feedback: {{feedback_file}}
Suggested fix: {{fix}}
### Recommendation
{{overall: ship / revise-minor / revise-major / block}}
### Candidate feedback to file
(optional, present only if the reviewer found a new failure pattern)
### Specialization signal
(optional, present only if the reviewer flags a domain that should graduate into its own specialized reviewer)Surface WARN items prominently even when FAIL is zero — they're the ones that graduate into feedback_*.md next session if ignored. The reviewer may also surface a "subtype auto-detected" note at the top if the dispatcher didn't supply one; pass it through unchanged.
If the reviewer returned no FAIL items, still surface WARN items prominently — they're the ones that graduate into feedback_*.md next session if ignored.
Reviewer Agent Contract
Each deliverable-*-reviewer agent must:
Load feedback memory dynamically. Glob
~/.claude/projects/<project-id>/memory/feedback_*.md. Filter to verification-relevant entries (those describing a past verification failure, scope error, missed exception, misattribution, or premature conclusion). Read the top 5–15 by relevance.Path resolution:
~은 사용자 홈($HOME또는$USERPROFILE)으로 자동 확장된다.<project-id>는 현재 Claude Code 세션의 project hash로, CWD 기반으로 생성되어 Claude Code가 환경에서 해석한다. auto-memory가 비어 있거나 접근 불가하면 레포 내부memory/feedback_*.md로 fallback (아래 "Failure Handling" 섹션 참조).- Merge static + dynamic checklist. Each agent has a domain-specific static checklist baked into its prompt. Merge it with the dynamic feedback-derived items. Deduplicate.
- Walk the file against the checklist. For each item: PASS / WARN / FAIL with evidence (file line or feedback file path).
- Name the feedback source. Every WARN or FAIL must cite which
feedback_*.mdor which static rule it came from. This is how 사용자 audits the reviewer itself. - Suggest a concrete fix for each FAIL. Not "revisit this section" — an actual edit direction.
- Flag new failure patterns. If the reviewer notices a failure mode that isn't yet in feedback memory, end the report with
### Candidate feedback to file— a short entry 사용자 can save as a newfeedback_*.md.
Growth Path
The general reviewer is the incubator. When the same failure class recurs 3+ times in general reviewer output, that class should graduate into its own specialized reviewer. 사용자 decides when to split.
Current roster:
deliverable-general-reviewer— fallback + incubatordeliverable-meddev-compliance-reviewer— 의료기기·법규·SOP·계약deliverable-spec-reviewer— 기술 spec·제안서·아키텍처 문서
Future candidates (do not build until pattern justifies):
deliverable-linkedin-reviewer— voice/tone check for 사용자's LinkedIn postsdeliverable-wiki-reviewer— cross-reference integrity, orphan detection
Red Flags — STOP and run the full gate
These thoughts mean you're rationalizing your way around the check:
| Thought | Reality |
| "This file is small, skip the gate" | Small files are where scope errors hide. The gate takes 60 seconds. |
| "I already read the source carefully" | Past-사용자 said that exact sentence before feedback_law_full_reading got filed. |
| "The wrap skill will catch it" | wrap delegates to this skill. There is no other net. |
| "General reviewer is generic, it won't find anything" | That's a signal to strengthen the general reviewer, not to skip it. |
| "사용자 said it's ready, they know best" | 사용자 is the one who built this skill precisely because "ready" has been wrong before. |
| "The deliverable is in a language the reviewer doesn't handle well" | Then warn about low confidence in the report — don't silently skip. |
No exceptions. If a file matches the trigger conditions, the gate runs. 사용자 can read a 10-line PASS report in 3 seconds; the cost of skipping is measured in new feedback entries.
Failure Handling
If the named reviewer agent is not yet registered in ~/.claude/agents/:
- Warn 사용자 once: "deliverable-{{type}}-reviewer not found, falling back to general-purpose agent with inline instructions"
- Pass the same contract (load feedback memory, static + dynamic checklist, evidence-cited output) as an inline prompt to the general-purpose subagent
- Note in the final report which reviewer was actually used
If the file path doesn't exist or is empty:
- Report and stop. Do not fabricate a review.
If feedback memory directory is missing:
- Proceed with static checklist only and note this degradation at the top of the report.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

