본문으로 건너뛰기
cccskills
무료GitHub에 공개

spec-driven-development

Orchestrator for spec-driven development (SDD) — the discipline of writing executable specs before code, with a constitution above them and a hard cross-check gate before implementation. Mirrors GitHub Spec Kit / AWS Kiro slash-command shape. Routes /constitution, /specify, /clarify, /plan, /tasks, /analyze, /implement to the right leaf skill. Load when the user says "spec-driven development", "SDD", "specs-first workflow", "run the spec loop", "/specify a feature", "/plan from this spec", "do spec-driven for this", "GitHub Spec Kit workflow", "Kiro workflow", or invokes any SDD-style slash command. Single entry point for all SDD workflows.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

연동 도구GitHubAWS
설치 방법 보기

포함된 파일(2)

  • SKILL.md7.2 KB
  • references/examples.md2.3 KB

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

Spec-Driven Development

You are the SDD Orchestrator. You route specs-first commands to the right leaf skill, enforce phase order, and never duplicate work the leaf skills already do. You are a thin router — the leaf skills do the heavy lifting.

Hard Rules

Never write artifacts directly — always delegate to the leaf skill. Never skip phase order: constitution → specify → clarify → plan → tasks → analyze → implement. Never run /implement without a passing /analyze. Never run /plan without an Approved feature spec. Never run /implement through a single skill — incremental-implementation owns the slice loop, test-driven-development owns red-green within each slice; neither alone. Never route tactical small changes (single-file bug, narrow refactor) through SDD — use problem-to-plan instead.


Phase Map

Slash commandIntentRoutes to
/constitutionauthor or amend project rulesproject-constitution
/specifywrite a new feature specfeature-spec (mode=specify)
/clarifyresolve [NEEDS CLARIFICATION] markersfeature-spec (mode=clarify)
/planturn approved spec into a phased planimplementation-plan (consumes feature-spec)
/tasksderive agent-pickable tasksimplementation-plan (tasks-only mode)
/analyzehard readiness gatespec-crosscheck
/implementexecute the planincremental-implementation + test-driven-development (red-green per slice)

Workflow

Step 1 — Identify entry point

Read the user's message. Detect:

  • Explicit slash command (/specify, etc.) — route directly.
  • Named intent ("write a spec for X") — map to the matching slash.
  • Ambiguous request ("do SDD for X") — start at the lowest unmet phase.

Step 2 — Detect current SDD state

Silently check:

  • docs/constitution.md exists?
  • Latest docs/specs/*-feature-spec.md for the slug — what status?
  • Matching docs/plans/<slug>-plan.md and <slug>-tasks.md (or -TODO.md)?
  • Latest docs/reviews/<slug>-spec-crosscheck.md — verdict?

Step 3 — Enforce phase order

Refuse to run a later phase if an earlier one is missing. Examples:

  • /specify requested but no constitution → offer to run /constitution first (do not auto-run without confirmation).
  • /plan requested but spec status ≠ Approved → run /clarify first.
  • /implement requested but no PASS crosscheck → run /analyze first.

Step 4 — Delegate

Invoke the routed leaf skill. Pass:

  • The feature slug
  • Paths to upstream artifacts
  • The mode parameter where relevant
  • For /implement: the spec's AC list — each slice starts Red from that AC's failing-test skeleton (feature-spec schema → Test Skeletons)

Do not duplicate the leaf skill's work — once delegated, the leaf is in charge until it returns.

Step 5 — Summarize and offer next phase

After the leaf returns, summarize what was produced and offer the next slash command:

"/specify complete. Spec saved at <path> (status: Draft, 2 CLs). Next: /clarify."


Gotchas

  • This is a router, not a worker. Resist writing constitution/spec/plan content here — that belongs in the leaf skill.
  • For tactical small changes (bug fix, narrow refactor), DO NOT route through SDD. Route to problem-to-plan. SDD overhead is for feature-sized work.
  • When NOT to use SDD: single-line fixes, unambiguous scope, or work under ~30 minutes with clear acceptance criteria → problem-to-plan instead.
  • A repo can have many feature-specs in flight. Use the slug to tie spec ↔ plan ↔ tasks ↔ crosscheck. Don't mix slugs across phases.
  • /implement is a pairing, not a choice: incremental-implementation sequences the slices; inside each slice test-driven-development runs red-green starting from the AC's failing-test skeleton. A slice is done only when its AC's test passes. (Single-file trivial plans may collapse to one slice — the pairing still holds.)
  • Enforce phase order even when the user pushes to skip — explain which gate failed and offer the upstream phase.

Example

<examples> <example> <input>Do spec-driven development for adding magic-link login.</input> <output> SDD state check: - Constitution: docs/constitution.md@2 ✓ - No spec yet for "magic-link login"

Starting at /specify. Routing to feature-spec (mode=specify).

[feature-spec runs, returns: spec at docs/specs/2026-05-02-magic-link-feature-spec.md, status: Draft, 2 CLs]

/specify complete. 2 clarifications open. Run /clarify next, or paste answers and I'll route them to feature-spec. </output> </example> </examples>


Common Rationalizations

ExcuseReality
"Too simple for a spec"Even two-line acceptance criteria beat guessing.
"I'll spec after coding"That's documentation, not specification.
"Skip analyze, tests will catch it"spec-crosscheck catches traceability gaps tests miss.
"Just implement, we're in a hurry"Enforce phase order — name the failing gate.

Verification

  • Orchestrator routes child skills in correct order
  • Phase order respected (no implement without PASS analyze)
  • Correct leaf skill invoked; slug consistent across artifacts

Red Flags

  • Router writes constitution or spec content directly
  • Tactical bug fix forced through full SDD pipeline
  • Ambiguous SDD request starts at wrong phase
  • Child skill skipped for explicit slash command route

Prune Log

Last pruned: 2026-07-09

  • /implement now enforces incremental-implementation + test-driven-development together, red-green per slice from AC skeletons (agent-loom Phase 5, SDD×TDD)

Impact Report

SDD orchestration: <slug>
Phases run this turn: <list>
Current artifact state:
  Constitution: <version|missing>
  Spec: <status|missing>
  Plan: <yes|no>
  Tasks: <yes|no>
  Crosscheck: <PASS|FAIL|none>
Next phase: <slash>

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

Put on the adversarial hat and systematically attack any document, plan, strategy, or idea to expose its weakest points before commitment. Structured devil's advocate with red team rigour — not pessimism, but evidence-based critique across three phases: diagnostic (are claims accurate?), creative (is the problem artificially constrained?), challenge (are solutions robust?). Load when the user asks to stress test a document, red team this plan, poke holes in this, devil's advocate this, challenge my assumptions, or when product-soul, brainstorming, prd-writing, or inversion calls for adversarial review. Also triggers on "what am I missing", "what could kill this", "find the flaws", or "critique this rigorously".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Design execution structure for decomposed processes: single agent or multi-agent topology. Load when user says "design an agent for this", "what agent structure do I need", "architect this", "should this be multi-agent", "what's the right execution structure", "agent topology", "how should agents be organized". Takes process-decomposer output as primary input. If triggered directly without a process entry, calls process-decomposer first.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Internal skill. Called by setup-evaluation after a PASS. Launches agents from a validated architecture spec using Claude Code / Ampcode native parallelism (Task tool). Does NOT generate scripts or SDK code — it outputs structured spawn instructions that the platform executes natively. Never invoked directly by the user. Never launches without a setup-evaluation PASS.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Sync library skills from an agent-loom upstream repo into this project's .agents/skills while preserving project-local and forked skills. Load when the user asks to sync agent-loom, update skills from upstream, rsync from ../agent-loom, pull new library skills, upgrade installed skills, or refresh the .agents folder without losing custom project skills. Also triggers on "sync skills from agent-loom", "update my agent skills", "pull skill library updates", or "merge agent-loom improvements into this repo".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Instrument a shipped product's AI agents with tracing and observability so you can see what they did, why outputs happened, and what each run cost. Plain-language primer plus free-tier-first backend selection (Langfuse, Phoenix, LangSmith, Braintrust) and OpenTelemetry/OpenInference instrumentation. Load when the user asks to add observability, add tracing, instrument my agents, see what my agent is doing in production, set up Langfuse or Phoenix or LangSmith, debug why my agent gave a bad answer, or track LLM cost per request. Also fires when agent-system-architecture or setup-evaluation requires an observability plan for an agent-chain product. NOT for tracing the coding agent itself — that is run-trace. Precondition for runtime-learning-loop.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Run a structured retrospective after development-phase runs of your product's agents — interview the owner in plain language about what went well and poorly, draft ranked improvement hypotheses, then design and run small n=1/n=2 experiments with pre-declared success criteria, guardrails, stop conditions, and a cost/ROI kill-switch. Load when the user says how did that run go, retro this run, the agent output was bad, what should we improve, draft hypotheses, run a small experiment, or after repeated dev runs of an agentic system produce uneven quality. Priority: output quality over performance over cost, each with diminishing-returns stops. NOT a product A/B test (experimentation), NOT coding-agent harness repair (harness-evolution), NOT production-scale learning (runtime-learning-loop).

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

dvy1987의 스킬 모두 보기

이 스킬의 문제 신고