Enhance your AI coding tool with optional project bootstrap and a structured, spec-driven feature-delivery workflow. Define your product vision, create reusable system design context when needed, then move from specification to QC with quality built in, not bolted on.
Projects can optionally bootstrap shared product vision and technical context first. Then every feature moves through enforced phases with dedicated agents, structured artifacts, and quality gates.
Define your product vision with /sddp-prd, establish a reusable technical baseline with /sddp-systemdesign, then enter a strict gated lifecycle. You still can't plan without a spec or implement without tasks.
Each phase has a dedicated persona — Product Strategist, Solution Architect, Product Manager, Software Architect, QA Engineer, QC Agent, Project Manager — plus sub-agents for delegation.
Tasks link to functional requirements ({FR-###}) and user stories ([US#]). Nothing gets lost between spec and code.
Five pattern-based strategies (Adverb Trap, Passive Voice, Unspecified Scale, Missing Failure Mode, Scope Creep Detector) catch vagueness before it becomes bugs.
"Unit Tests for English" — checklists that test requirements quality, not code behavior. Auto-evaluated with PASS / RESOLVE / ASK outcomes.
Interrupted? Re-run /sddp-implement — completed tasks are automatically skipped. Pick up where you left off.
Shared, tool-agnostic skill files power GitHub Copilot, Claude Code, Windsurf, Antigravity, OpenCode, OpenAI Codex, and Gemini CLI through thin wrapper layers.
Cross-artifact consistency checks: spec vs. plan alignment, plan vs. tasks alignment, project instructions compliance — up to 50 findings with severity levels.
A dedicated Technical Researcher agent fetches best practices, official docs, and industry standards from the web — then distills findings into a concise research.md so your plan is evidence-backed.
Run the entire feature-delivery pipeline with /sddp-autopilot. Bootstrap stays explicit, while downstream phases auto-resolve decisions and record every choice.
Projects can optionally bootstrap shared product and architecture context before the strict feature-delivery lifecycle begins. Each feature phase still produces structured artifacts and must pass quality gates.
Dashed lines = optional phases
Dashed lines = optional / alternative paths
Product Strategist creates or refines specs/prd.md as the canonical Product Requirements Document and registers it in .github/sddp-config.md.
Solution Architect creates or refines specs/sad.md as the canonical technical context and registers it in .github/sddp-config.md.
DevOps Strategist creates or refines specs/dod.md as the canonical Deployment & Operations Document and registers it in .github/sddp-config.md.
Project Planner creates or refines specs/project-plan.md as the canonical Project Implementation Plan and registers it in .github/sddp-config.md.
Project Initializer establishes project-instructions.md and preserves or adopts the registered Product Document, Technical Context Document, Deployment & Operations Document, and Project Plan in .github/sddp-config.md.
Project Amender propagates a project-level change across project-instructions.md, the canonical PRD, SAD, DOD, and project plan by analyzing impact and inline-executing the owning bootstrap workflows.
Product Manager agent captures WHAT users need and WHY. Produces spec.md with prioritized user stories and functional requirements.
Business Analyst agent detects ambiguity using 5 pattern-based strategies. Asks targeted questions and encodes answers directly into the spec.
Software Architect agent designs the technical approach. Produces plan.md, research.md, data models, and API contracts while reusing the registered technical context when present.
QA Engineer agent generates domain-specific quality checklists in checklists/*.md. Covers testing, security, accessibility, and more.
Project Manager agent decomposes spec + plan into dependency-ordered tasks.md. Each task is atomic, testable, and traceable to requirements.
Compliance Auditor agent runs cross-artifact traceability analysis. Detects gaps, contradictions, and missing coverage between spec, plan, and tasks.
Software Engineer agent executes tasks phase-by-phase. Completed tasks marked [X] are skipped on re-run. Fully resumable.
QC Agent runs tests, static analysis, security audits, and story verification. Failures loop back as [BUG] tasks. Pass creates a .qc-passed marker.
Project bootstrap writes shared artifacts outside feature folders. Feature delivery then produces a complete set of traceable documents per feature, from requirements through QC.
specs/
├── prd.md # Canonical product requirements
├── sad.md # Canonical technical context
├── dod.md # Canonical deployment & operations
└── project-plan.md # Canonical project implementation plan
project-instructions.md # Project governance
.github/
└── sddp-config.md # Shared document registration + autopilot settings
specs/00001-user-auth/
├── spec.md # Feature specification
├── plan.md # Implementation plan
├── tasks.md # Phased task list
├── research.md # Technology research
├── data-model.md # Entity definitions
├── quickstart.md # Integration guide
├── contracts/ # API contracts
│ └── auth-api.md
├── checklists/ # Quality checklists
│ ├── spec-quality.md
│ └── plan-quality.md
├── .completed # Implementation done marker
├── autopilot-log.md # Autopilot audit log
└── .qc-passed # QC gate passed marker
Up and running in under 5 minutes. Bootstrap shared technical context when you need it, then start feature delivery with the same command sequence across supported tools.
Download the release archive for your AI tool and extract it into your project.
# Download a tool-specific release
gh release download --repo attilaszasz/sdd-pilot \
--pattern "sdd-pilot-copilot-*.zip" --dir .
unzip sdd-pilot-copilot-*.zip -d my-project
Transform product ideas into a structured specs/prd.md that provides downstream phases with product vision, user personas, and domain context.
# Create or refine specs/prd.md and register it
/sddp-prd Here are my product ideas, target audience,
and key features I want to build
Generate or refine specs/sad.md when you want a reusable architecture baseline before feature work.
# Create or refine specs/sad.md and register it
/sddp-systemdesign Use the attached PRD, architecture notes,
and deployment constraints to create the canonical specs/sad.md
Creates or refines the Deployment & Operations Document (specs/dod.md). Uses shared architecture and product context to define environments, CI/CD, and operational requirements.
# Define production environment requirements
/sddp-devops Define production environment requirements,
deployment topology, and CI/CD pipelines.
Creates or refines the Project Implementation Plan (specs/project-plan.md). Decomposes the product features into prioritized, dependency-ordered epics guiding the delivery schedule.
# Create an implementation plan
/sddp-projectplan Create an implementation plan based on the
system design and product requirements.
Define the non-negotiable principles that all downstream agents will enforce while preserving the registered technical context.
# Establish project-instructions.md and preserve shared context
/sddp-init Initialize project governance using attached PRD
Start feature delivery with a natural-language description, then remove ambiguity before planning.
# Create a feature branch
git checkout -b 00001-user-auth
# Specify what you want to build
/sddp-specify Build user authentication with email/password,
including registration, login, and password reset.
# Clarify any ambiguity
/sddp-clarify
Walk through the remaining phases. Each command produces structured artifacts and enforces quality gates.
# Design the technical approach
/sddp-plan
# Generate quality checklists (optional but recommended)
/sddp-checklist
# Decompose into dependency-ordered tasks
/sddp-tasks
# Run compliance analysis (optional but recommended)
/sddp-analyze
# Execute implementation — fully resumable
/sddp-implement
# Run quality control — tests, security, story verification
/sddp-qc
# Or loop implement + QC automatically (max 10 cycles)
/sddp-implement-qc-loop
Autopilot alternative: Run /sddp-autopilot Build user auth to execute the feature-delivery pipeline unattended. Bootstrap remains explicit.
Seventeen commands across project bootstrap and feature delivery. Every command maps to a specialized AI agent.
| Command | Agent | Produces |
|---|---|---|
| /sddp-prd | Product Strategist | specs/prd.md |
| /sddp-systemdesign | Solution Architect | specs/sad.md |
| /sddp-devops | DevOps Strategist | specs/dod.md |
| /sddp-projectplan | Project Planner | specs/project-plan.md |
| /sddp-amend | Project Amender | Updated bootstrap artifacts |
| /sddp-init | Project Initializer | project-instructions.md |
| /sddp-devsetup | Environment Setup Analyst | Local dev environment |
| /sddp-specify | Product Manager | spec.md |
| /sddp-clarify | Business Analyst | Updated spec.md |
| /sddp-plan | Software Architect | plan.md, research.md |
| /sddp-checklist | QA Engineer | checklists/*.md |
| /sddp-tasks | Project Manager | tasks.md |
| /sddp-analyze | Compliance Auditor | Markdown report |
| /sddp-implement | Software Engineer | Source code |
| /sddp-qc | QC Agent | .qc-passed / [BUG] tasks |
| /sddp-implement-qc-loop | Software Engineer (looping) | Source code + .qc-passed |
| /sddp-autopilot | Pipeline Orchestrator | All artifacts + autopilot-log.md |
One set of shared skills powers every supported AI tool. Each tool gets a thin wrapper layer that translates its native format into the shared skill engine.
All methodology logic lives in .github/skills/ — tool-agnostic Markdown files that define how each SDD phase works. Write once, run everywhere.
Each AI tool gets a minimal wrapper in its native format — Copilot prompts, Antigravity workflows, Claude agents, Windsurf workflows, OpenCode commands, Codex skills, and Gemini extensions — that delegates to the shared skills.
Sub-agents like the QC Auditor, Story Verifier, and Technical Researcher are defined in .github/agents/ and invoked identically by every tool.
Every release builds seven separate zip archives — one per tool. Each contains only the files that tool needs: its wrappers plus the shared core.
Because every tool reads the same skill files, you get identical phase gates, artifact formats, and quality standards — regardless of which AI tool you choose.
Adding a new AI tool means writing a thin wrapper layer. The skill engine, sub-agents, and governance rules are already there and ready to use.
sdd-pilot/
├── .github/ # Shared core (all tools read from here)
│ ├── skills/ # Shared skills, including system-design
│ ├── agents/ # Sub-agent definitions + Copilot agents
│ ├── instructions/ # Governance rules
│ ├── prompts/ # Copilot prompt wrappers
│ └── sddp-config.md # Product / technical context registration
│
├── specs/
│ ├── prd.md # Canonical product requirements document
│ ├── sad.md # Canonical technical context document
│ ├── dod.md # Canonical deployment & operations
│ └── project-plan.md # Canonical project implementation plan
├── .agents/workflows/ # Antigravity workflow wrappers
├── .windsurf/workflows/ # Windsurf workflow wrappers
├── .claude/agents/ + skills/ # Claude Code agent + skill wrappers
├── .opencode/agents/ + cmds/ # OpenCode agent + command wrappers
├── .codex/agents/ # Codex custom agent definitions
├── .agents/skills/ # Codex skill entry points
│
├── project-instructions.md # Project governance (shared)
└── AGENTS.md # Agent conventions (shared)