SDD Framework: The Spec is the Solution.
DeepSpec is a zero-ceremony, AI-native Spec-Driven Development (SDD) framework. It guides AI agents from intention to implementation using a strict 3-stage pipeline and the A-B-C documentation flow.
npx deep-spec init cursor-agentThen in your agent:
"Initialize DeepSpec"— generateAGENTS.mdfrom your repo"Create task [name]"— draft A-B-C specs (no app code)"Approve task"— implement with TDD in the same turn- Review at Review Gate, then
"Complete task"to archive
drafts/ → (Approve) → active/ → Review Gate → (Complete) → archive/
| Doc | Role |
|---|---|
APPROACH.md |
How — immutable execution plan |
BUSINESS_CONTEXT.md |
Why — acceptance criteria for TDD |
COMPLETION_REPORT.md |
Evidence — status and execution log |
CLI (recommended):
npx deep-spec init <agent>Available agents: cursor-agent, claude, copilot, opencode, cline, and more — run npx deep-spec --help.
Manual skill install (retrocompat):
mkdir -p .cursor/skills/deep-spec
cp -R deep-spec/. .cursor/skills/deep-spec/| Command | Trigger |
|---|---|
/deepspec.init |
Initialize DeepSpec |
/deepspec.create-task |
Create draft task |
/deepspec.approve-task |
Approve + execute |
/deepspec.discard-task |
Discard draft |
/deepspec.complete-task |
Archive after Review Gate |
/deepspec.revise-task |
Review Rounds |
/deepspec.list |
List tasks by stage |
/deepspec.repair |
Repair tracking map |
/deepspec.map-codebase |
Refresh AGENTS.md from repo scan |
/deepspec.diagram-architecture |
Mermaid ARCHITECTURE.md per task |
/deepspec.interview |
One-question approach refinement |
/deepspec.answer-questions |
Incorporate open-question answers |
Optional per-task artifact: ARCHITECTURE.md (diagrams on demand). The VS Code extension shows Execution Plan progress from APPROACH.md reconciled with COMPLETION_REPORT.md.
spec/commands/ # operational prompts
spec/templates/ # A-B-C templates
src/ # CLI + hooks (TypeScript)
deep-spec/ # orchestrator skill
.deepspec/ # dogfooded instance
MIT — see LICENSE.
