Pre-action boundary checking — validates agent tool calls against declared capabilities and task contracts
日本語の概要は準備中です。原文の説明を表示しています。
Spec-Driven Development workflow — enforces sdd/ folder hierarchy with planning-first gates, current-state artifacts, and completion verification
インストール方法を見るインストールする前に、エージェントに与えられる指示の中身を確認できます。
Spec-Driven Development workflow that enforces the sdd/ folder hierarchy. All development work proceeds through planning-first gates and produces current-state artifacts as completion evidence.
Invoke this skill when user requests:
sdd/
├── 01_planning/ # Requirements, constraints, stakeholder input
├── 02_plan/ # Execution plan, acceptance criteria, approach
├── 03_build/ # Current build state, implementation notes
├── 04_verify/ # Verification evidence, test results, residual risks
├── 05_operate/ # Deployment notes, runbooks (conditional)
├── 99_toolchain/ # Tool configs, scripts, environment setup
└── decisions/ # Decision records for major design choices
Key Principle: These folders are current-state artifacts, not history archives. Each file reflects the current state of the work — update in place rather than appending new versions.
Verify sdd/ folder exists in the project root:
ls sdd/ 2>/dev/null || echo "sdd/ folder not found"
If sdd/ does not exist:
sdd/ foldermkdir -p sdd/{01_planning,02_plan,03_build,04_verify,05_operate,99_toolchain,decisions}If sdd/ exists, continue to Step 1.
MUST complete before any coding begins.
Read all relevant planning documents:
[SDD] Reading planning documents...
├── sdd/01_planning/ — requirements, constraints
└── sdd/99_toolchain/ — tool configs, environment
Actions:
sdd/01_planning/ contents (all .md files)sdd/99_toolchain/ if presentGate: If planning docs are absent or insufficient, pause and ask user to provide requirements. Do NOT proceed to plan phase without understanding the context.
Create or update sdd/02_plan/ with the execution plan.
Artifact to produce: sdd/02_plan/current.md
# Execution Plan
## Task
{task description}
## Approach
{implementation strategy}
## Acceptance Criteria
- [ ] {criterion 1}
- [ ] {criterion 2}
- [ ] {criterion N}
## Out of Scope
- {what will NOT be done}
## Dependencies
- {prerequisite or dependency}
Display:
[SDD Plan] Created sdd/02_plan/current.md
├── Approach: {one-line summary}
├── Criteria: {N} acceptance criteria defined
└── Gate: Plan ready — proceeding to build
Implement the changes. Update sdd/03_build/ with current build state.
Artifact to produce or update: sdd/03_build/current.md
# Build State
## Status
{In Progress | Complete}
## Implemented
- {file or component}: {what was done}
## Decisions Made
- {decision}: {rationale}
- Write DR for major decisions: sdd/decisions/{YYYY-MM-DD}-{topic}.md (template: templates/decision-record.md)
## Known Issues
- {issue}: {planned resolution}
During implementation:
sdd/03_build/current.md as work progressesDisplay:
[SDD Build] Implementing changes...
[SDD Build] Updated sdd/03_build/current.md
└── Status: {In Progress | Complete}
Run verification and update sdd/04_verify/ with evidence.
Artifact to produce or update: sdd/04_verify/current.md
# Verification Evidence
## Acceptance Criteria Results
- [x] {criterion 1}: {evidence}
- [x] {criterion 2}: {evidence}
- [ ] {criterion N}: {reason if not passing}
## Tests Run
- {test command or suite}: {result}
## Residual Risks
- {risk}: {severity} — {mitigation plan}
## Verdict
{Pass | Fail | Conditional Pass}
Actions:
sdd/02_plan/current.mdGate: If verdict is Fail, return to Step 3 (Build). Do NOT declare done with a Fail verdict.
Display:
[SDD Verify] Running verification...
[SDD Verify] Updated sdd/04_verify/current.md
├── Criteria: {N}/{total} passing
├── Residual risks: {count}
└── Verdict: {Pass | Fail | Conditional Pass}
Only execute if deployment is in scope for this task.
Artifact to produce or update: sdd/05_operate/current.md
# Operate State
## Deployment
- Environment: {target}
- Method: {how deployed}
- Timestamp: {when}
## Runbook
{steps to operate, restart, rollback}
## Monitoring
{what to watch, alerts, logs}
Skip this step if:
Before declaring [Done], verify:
[SDD Done?] Checking completion gates...
├── sdd/02_plan/current.md — exists? {Y/N}
├── sdd/03_build/current.md — exists and current? {Y/N}
├── sdd/04_verify/current.md — exists and verdict Pass? {Y/N}
└── sdd/05_operate/current.md — exists (if deploy in scope)? {Y/N}
If any gate fails, complete the missing artifact before declaring done.
Final display:
[SDD Done] Task complete
├── Plan: sdd/02_plan/current.md
├── Build: sdd/03_build/current.md
├── Verify: sdd/04_verify/current.md (Verdict: Pass)
└── Artifacts are current-state — ready for next iteration
current.md is the canonical artifact. Supplementary files are allowed but the main doc is always current.md.Major design decisions during Step 3 are recorded in sdd/decisions/{YYYY-MM-DD}-{topic}.md using the template at templates/decision-record.md (relative to this skill directory).
When to create a Decision Record:
See guides/harness-engineering/ for harness-level decision context that DRs may reference.
| Skill | When to Use Together |
|---|---|
/deep-plan | Before /sdd-dev when requirements are ambiguous — use deep-plan to research, then sdd-dev to execute |
/structured-dev-cycle | Alternative workflow without sdd/ folder — use sdd-dev when project uses sdd/ hierarchy |
/dev-review | After Step 4 — use dev-review for additional code quality check |
/adversarial-review | After Step 4 — use adversarial-review for security-sensitive changes |
/sdd-dev add user authentication to the API
/sdd-dev implement pagination for the product list screen
/sdd-dev 화면명세서 기반으로 대시보드 UI 구현
/sdd-dev refactor the payment module
/sdd-dev # (no argument — guided workflow)
When no argument is provided, ask: "어떤 작업을 진행할까요? sdd/01_planning/ 의 요구사항을 기반으로 계획을 수립합니다."
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Pre-action boundary checking — validates agent tool calls against declared capabilities and task contracts
日本語の概要は準備中です。原文の説明を表示しています。
Auto-detect project context and optimize harness — deactivate unused agents/skills, suggest missing experts, generate project profile
日本語の概要は準備中です。原文の説明を表示しています。
Adversarial code review using attacker mindset — trust boundary, attack surface, business logic, and defense evaluation
日本語の概要は準備中です。原文の説明を表示しています。
Apache Airflow best practices for DAG authoring, testing, and production deployment
日本語の概要は準備中です。原文の説明を表示しています。
Alembic migration patterns for naming conventions, safety checks, expand-contract, env.py configuration, and CI integration
日本語の概要は準備中です。原文の説明を表示しています。
Pre-routing ambiguity analysis — scores request clarity and asks clarifying questions when needed (inspired by ouroboros)
日本語の概要は準備中です。原文の説明を表示しています。