curl -L -o spec-interview-playbook.skill "https://aiskillstore.io/v1/agent/skills/bf5c84ab-0857-4701-8d56-7f9eeff62131/download?platform=ClaudeCode"
{
"tool": "download_skill",
"arguments": {
"skill_id": "bf5c84ab-0857-4701-8d56-7f9eeff62131",
"platform": "ClaudeCode"
}
}
{
"mcpServers": {
"skill-store": {
"url": "https://aiskillstore.io/mcp/"
}
}
}
Playbook: turn a vague request into a decision log, spec and dependency-ordered tickets. Use when clarifying requirements before coding, interviewing a requester one question at a time with recommended answers, or writing a spec with testable acceptance and assumption kill criteria. 모호한 요구를 질문트리로 사양·티켓화하는 플레이북.
Compatible Platforms any
Findings: ["메타데이터 경고: 권장 필드 없음: 'requirements' (SKILL.md v2 권장)", "메타데이터 경고: 권장 필드 없음: 'changelog' (SKILL.md v2 권장)"]
✅ No security risks found.
AI Review Stage
스킬 메타데이터에 따르면, 이 스킬은 네트워크, 파일 시스템, 서브프로세스 접근 권한이 없다고 명시되어 있습니다 (`'network': false`, `'filesystem': false`, `'subprocess': false`). 코드 파일을 분석한 결과, 이러한 선언과 일치하는 것을 확인했습니다. - **main.py**: `os.path.dirname(os.path.abspath(__file__))`와 같은 `os` 모듈 사용은 내부 모듈 로딩 경로 설정에만 사용되며, 사용자 파일 시스템에 대한 읽기/쓰기 작업을 수행하지 않습니다. - **lib/common.py**: `emit` 함수는 `sys.stdout.buffer.write`를 통해 표준 출력으로만 데이터를 내보내며, `read_payload` 함수는 `sys.stdin.buffer.read`를 통해 표준 입력에서만 데이터를 읽습니다. `file_entry` 함수는 파일 내용을 메모리 상의 딕셔너리 형태로 생성할 뿐, 실제 디스크에 파일을 쓰지 않는다고 명확히 설명되어 있습니다. `re` 모듈은 문자열 처리에만 사용됩니다. - **lib/spec.py**: 이 모듈은 주로 질문 목록, 규칙, 문서 생성 로직을 포함하며, 모두 인메모리 데이터 처리 및 문자열 생성에 국한됩니다. 코드 전체에서 외부 네트워크 통신(예: `requests`, `urllib`), 사용자 파일 시스템 접근(예: `open()`, `shutil`), 서브프로세스 실행(예: `subprocess`, `os.system`) 등의 위험한 기능 사용 흔적은 발견되지 않았습니다. 또한, 코드 난독화나 악의적인 데이터 탈취/시스템 파괴를 목적으로 하는 코드는 없었습니다. 사용자 데이터를 무단으로 수집하거나 전송하는 기능도 없습니다. 스킬의 목적(요구사항 명확화, 사양 및 티켓 생성)과 코드 품질은 일치하며, 정적 분석 결과도 'approved' 상태로 위험 요소가 없음을 확인했습니다.
Representative input/output examples for this skill. Agents can use these to understand how to invoke the skill and what output to expect.
Ask one question at a time. Q-GOAL and Q-USERS are already answered, so the next P0 question is scope. (output abridged)
{
"action": "question_plan",
"answered": [
"Q-GOAL",
"Q-USERS"
],
"topic": "Team invite links"
}
{
"action": "question_plan",
"ask_next": {
"answered": false,
"category": "scope",
"codebase_probe": "Look for TODOs and open issues that overlap.",
"id": "Q-SCOPE",
"priority": "P0",
"question": "What is explicitly OUT of scope for \u0027Team invite links\u0027 in this iteration?",
"recommended_answer": "Name 2-3 tempting extras and defer them explicitly.",
"why": "Non-goals prevent silent scope growth and shorten the ticket list."
},
"lang": "en",
"p0_remaining": 3,
"playbook": "spec-interview-playbook",
"queue": [
{
"answered": true,
"category": "goal",
"codebase_probe": "Check the issue/PRD/README for an existing statement of the goal.",
"id": "Q-GOAL",
"priority": "P0",
"question": "What outcome must exist when \u0027Team invite links\u0027 is done, and who will notice it first?",
"recommended_answer": "One measurable outcome. If there are several, rank them and drop the lowest.",
"why": "Everything else is judged against this outcome."
},
{
"answered": true,
"category": "users",
"codebase_probe": "Search for current callers / existing workarounds in the code.",
"id": "Q-USERS",
"priority": "P0",
"question": "Who are the users or callers of \u0027Team invite links\u0027, and what do they do today instead?",
"recommended_answer": "Primary user = the one who feels the pain; list at most 2 user types for v1.",
"why": "Defines who acceptance is for and what is replaced."
},
{
"answered": false,
"category": "scope",
"codebase_probe": "Look for TODOs and open issues that overlap.",
"id": "Q-SCOPE",
"priority": "P0",
"question": "What is explicitly OUT of scope for \u0027Team invite links\u0027 in this iteration?",
"recommended_answer": "Name 2-3 tempting extras and defer them explicitly.",
"why": "Non-goals prevent silent scope growth and shorten the ticket list."
},
{
"answered": false,
"category": "constraints",
"codebase_probe": "Read package manifests, CI config and deployment docs.",
"id": "Q-CONSTRAINTS",
"priority": "P0",
"question": "Which constraints are fixed for \u0027Team invite links\u0027 (deadline, stack, compatibility, compliance, budget)?",
"recommended_answer": "Keep the current stack and public interfaces unchanged unless a requirement forces otherwise.",
"why": "Constraints eliminate options before you discuss them."
},
{
"answered": false,
"category": "acceptance",
"codebase_probe": "Find the existing test style and reuse its fixtures.",
"id": "Q-ACCEPT",
"priority": "P0",
"question": "Which 3 checks would prove \u0027Team invite links\u0027 works? Each must be testable by a command or a clear observation.",
"recommended_answer": "Write them as Given / When / Then with concrete values.",
"why": "Untestable acceptance criteria become opinion fights later."
},
{
"answered": false,
"category": "data",
"codebase_probe": "Inspect the schema/migrations and sample records.",
"id": "Q-DATA",
"priority": "P1",
"question": "What data does \u0027Team invite links\u0027 read or write, who owns it, and how much is there?",
"recommended_answer": "Reuse existing tables/schemas; no new storage unless volume or lifecycle requires it.",
"why": "Data shape, volume and privacy drive the design."
},
{
"answered": false,
"category": "interfaces",
"codebase_probe": "List current endpoints/commands and their consumers.",
"id": "Q-INTERFACE",
"priority": "P1",
"question": "Which interfaces does \u0027Team invite links\u0027 touch (API, UI, CLI, events) and what must stay backward compatible?",
"recommended_answer": "Additive changes only; version anything breaking.",
"why": "Interface contracts create the most expensive rework."
},
{
"answered": false,
"category": "behavior_edge",
"codebase_probe": "Look at how similar features handle errors.",
"id": "Q-EDGE",
"priority": "P1",
"question": "What should happen on failure, empty input, duplicates, concurrency and missing permissions?",
"recommended_answer": "Fail loudly with a clear error; idempotent where retries are possible.",
"why": "Edge behavior is where specs usually stay silent and bugs start."
},
{
"answered": false,
"category": "quality",
"codebase_probe": "Check existing SLOs, lint and security rules.",
"id": "Q-QUALITY",
"priority": "P1",
"question": "What are the targets for performance, security, accessibility and observability?",
"recommended_answer": "Adopt the project\u0027s existing budgets; add one metric and one alert.",
"why": "Non-functional targets must be numbers to be testable."
},
{
"answered": false,
"category": "risks",
"codebase_probe": "Search past incidents and postmortems for similar work.",
"id": "Q-RISK",
"priority": "P1",
"question": "What are the top assumptions behind \u0027Team invite links\u0027, and what observation would make us stop or change course?",
"recommended_answer": "Record 2-3 assumptions in assumptions.md, each with a kill criterion and a check date.",
"why": "Kill criteria turn hidden assumptions into checkable bets."
},
{
"answered": false,
"category": "rollout",
"codebase_probe": "Read the release / deployment checklist.",
"id": "Q-ROLLOUT",
"priority": "P2",
"question": "How will \u0027Team invite links\u0027 be released and rolled back (flag, migration, owner, monitoring)?",
"recommended_answer": "Ship behind a flag with a one-step rollback.",
"why": "Release mechanics decide how risky the change really is."
}
],
"remaining": 9,
"rules": [
"Ask ONE question per turn.",
"Always attach your recommended answer and why; the human only has to confirm or correct.",
"Investigate the codebase and docs first; never ask what you can read (run the codebase_probe).",
"Prefer closed questions with 2-3 concrete options.",
"Record every answer immediately as a decision (decisions.md) or an assumption (assumptions.md).",
"Stop asking when remaining open items would not change the ticket list."
],
"stop_asking": false,
"topic": "Team invite links",
"version": "1.0.0"
}
Open decision D-002 is listed first; tickets come out in three dependency waves; the findings show what still blocks the exit check. (output abridged)
{
"action": "scaffold",
"state": {
"acceptance": [
"Given an admin, when they create a link, then it works for 7 days",
"The UI is user-friendly"
],
"assumptions": [
{
"check_by": "2026-12-01",
"kill_criterion": "p95 above 500 links/day in the first month",
"text": "Admins create fewer than 50 links per day"
},
{
"text": "Email delivery is not required"
}
],
"constraints": [
"Reuse the existing auth service",
"Ship behind a feature flag"
],
"decisions": [
{
"context": "Links must be unguessable and revocable.",
"decision": "Random DB token",
"id": "D-001",
"options": [
{
"cons": "hard to revoke",
"name": "Signed JWT",
"pros": "stateless"
},
{
"cons": "needs a table",
"name": "Random DB token",
"pros": "revocable"
}
],
"rationale": "Revocation is a hard requirement.",
"recommendation": "Random DB token",
"status": "accepted",
"title": "Link token format"
},
{
"id": "D-002",
"status": "open",
"title": "Default expiry"
}
],
"goal": "Let workspace admins invite teammates with a shareable link that expires.",
"non_goals": [
"SSO provisioning",
"Bulk CSV import"
],
"tickets": [
{
"acceptance": [
"dialog shows the link"
],
"blocked_by": [
"T-02"
],
"id": "T-03",
"title": "UI: invite dialog"
},
{
"id": "T-01",
"title": "Invite token table and migration"
},
{
"acceptance": [
"returns 201 with a token"
],
"blocked_by": [
"T-01"
],
"id": "T-02",
"title": "POST /invites endpoint"
}
],
"title": "Team invite links",
"users": [
"Workspace admin"
]
}
}
{
"action": "scaffold",
"files": [
{
"content": "# Decision log: Team invite links\n\nADR-style entries (Context / Options / Decision / Consequences). Open decisions come first.\n\n## D-002: Default expiry\n- Status: open\n- Context: \u003cto be filled\u003e\n- Options:\n - \u003coption A: pros / cons\u003e\n...",
"description": "Decision ledger in ADR style (open decisions first).",
"path": "decisions.md"
},
{
"content": "# Assumption ledger: Team invite links\n\nEvery assumption needs a kill criterion: the observation that makes us stop or change course.\n\n| ID | Assumption | Evidence | Risk if wrong | Kill criterion | Check by |\n|",
"description": "Assumptions with kill criteria.",
"path": "assumptions.md"
}
],
"lang": "en",
"playbook": "spec-interview-playbook",
"version": "1.0.0"
}
All examples are also available via the agent API:
/v1/agent/skills/bf5c84ab-0857-4701-8d56-7f9eeff62131/schema
No reviews yet. Be the first to leave one!