spec-kit (GitHub)
GitHub’s open-source (MIT, Python) toolkit for spec-driven-development — treating specifications as executable blueprints that “directly generate working implementations,” not static docs. The canonical SDD toolkit (already compared in spec-driven-ai-tools, now ingested as a primary source).
The workflow (six slash commands)
/speckit.constitution (project principles/governance) → /speckit.specify (requirements/user
stories) → /speckit.clarify (resolve ambiguities) → /speckit.plan (technical strategy) →
/speckit.tasks (ordered, actionable tasks) → /speckit.implement (execute). Plus GitHub-issue
creation and quality validation. Describe what & why before how.
Integration & design
- Works with 30+ AI coding agents (Copilot, Claude Code, Gemini, Cursor, …) via either slash-command prompts or agent-native skills (agent-skills).
- Modular: extensions (new capabilities) + presets (workflow tweaks); tool-agnostic, so teams can enforce org standards across agents.
Why it belongs here
The reference implementation of spec-driven-development and a clean example of process-as- markdown delivered as cross-agent skills/commands — the authoring-time discipline that agentic-coding-harness products wrap around. Greenfield + brownfield + iterative use.
Related
spec-driven-development · spec-driven-ai-tools · agent-skills · agentic-coding-harness