一键导入
specky-sdd-pipeline
Use for Specky contract architecture, lifecycle/workload/mode selection, feature state, phase transitions, tool scope, evidence, and gates.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use for Specky contract architecture, lifecycle/workload/mode selection, feature state, phase transitions, tool scope, evidence, and gates.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
This skill should be used when the user asks to 'research codebase', 'scan tech stack', 'import documents', 'discovery questions', or needs guidance on Phase 1 research. Also trigger on 'sdd research', 'brownfield scan', 'import transcript', or 'check ecosystem'.
Use for Phase 2 (Specify): write SPECIFICATION.md with EARS, REQ-IDs, and acceptance criteria. Trigger on sdd_write_spec, sdd_turnkey_spec, sdd_validate_ears, or /specky-specify.
Use for Phase 4 (Design): produce DESIGN.md with architecture, API contracts, data model, and Mermaid diagrams. Trigger on sdd_write_design, sdd_generate_all_diagrams, or /specky-design.
Use for contracted implementation planning, executable TDD bindings, IaC, Docker, devcontainer, local environment, or Codespaces tools after an APPROVE analysis gate.
Use when starting or resuming Specky work, selecting lifecycle/workload/mode/capabilities, diagnosing config/state, or choosing an agent/tool.
This skill should be used when the user asks to 'orchestrate pipeline', 'run full pipeline', 'coordinate phases', 'advance phase', or needs guidance on end-to-end pipeline execution. Also trigger on 'specky orchestrate', 'phase transition', 'LGTM gate', 'pipeline status', or 'which agent for this phase'.
| name | specky-sdd-pipeline |
| description | Use for Specky contract architecture, lifecycle/workload/mode selection, feature state, phase transitions, tool scope, evidence, and gates. |
Every feature is explicit: spec_dir plus a three-digit feature_number. Canonical state is signed and stored at .specs/NNN-name/.sdd-state.json. There is no root state, first feature, latest feature, 001, or .specs fallback in MCP schemas.
State embeds a versioned use-case snapshot and SHA-256 fingerprint. Tools reject identity mismatch, disabled capabilities, config-disabled contracts, invalid signatures, and legacy state.
All 54 lifecycle/workload/mode combinations are named contracts. Capabilities and parameter objects are additional explicit dimensions.
The signed contract.phases array controls transitions. Full includes Init, Discover, Specify, Clarify, Design, Tasks, Analyze, Implement, Verify, Release. Rapid and emergency persist smaller graphs. Never impose a global phase list; route only phases present in state.
Every MCP tool has one declared scope: stateless, workspace, bootstrap, batch, or feature. Feature tools require canonical state. Bootstrap/batch/workspace capability tools require an explicit use-case selection. Unknown tools have no compatibility fallback and fail registration/lookup.
[TODO:] substitution exists.Analyze computes APPROVE, CHANGES_NEEDED, or BLOCK. Implementation-sensitive tools centrally require APPROVE. Rewriting Specification, Design, or Tasks invalidates prior approval. LGTM blocking is controlled by strict workspace config.
Capability configuration is fingerprinted with state and cannot be overridden per call. Examples:
Use specky migrate-contracts --spec-dir=<path> --dry-run with explicit use-case mapping. Review the deterministic plan hash, then apply the identical plan with --confirm-plan. Migration backs up metadata, signs v5 state, verifies non-state artifact hashes, and rolls back on failure.
Agents are lean workflow routers. Their first action is to read the companion skill, which owns exact schemas, rules, examples, and evidence standards.