sddadd

The SDD + ADD Starter Template is a governance framework for AI programming assistants, particularly Claude Code. Its primary goal is to transform the AI assistant into a disciplined development partner that works according to a Spec-Driven Development (SDD) workflow combined with Agent-Driven Development (ADD). ## Core Workflow The template applies a sequential development workflow: CONTEXT → SPEC → PLAN → TASKS → execute → verify → sync. Each step has a checkpoint and requires human approval before the agent is allowed to execute. ## Three Foundational Rules 1. **Edit the Spec, not the Code**: When a requirement is missing or ambiguous, use the `/sdd-update` command to update the spec first, then review it before changing behavior. 2. **No Technical Speculation**: Plan, Tasks, and execution use only the bindings and exact verification commands approved in `.sdd/architecture-profile.md`. 3. **Humans Decide**: The agent does not self-approve, does not `git push`, does not deploy, and does not autonomously perform physical state changes. ## Ownership and Execution The template separates two independent axes in `.sdd/shared_context.md`: - **Project Ownership** (`solo` or `team`): Determines the validity of human review and Git delivery policy. Defaults to `team`. - **Agent Execution** (`direct` or `orchestrated`): `/add-execute` runs directly in the current agent or orchestrates workers after observing runtime capabilities. ## Documentation The template provides a comprehensive documentation system: - **Quick Start** (`docs/sdd-add-quickstart.md`): Step-by-step roadmap for newcomers. - **Operations Guide** (`docs/sdd-add-guide.md`): Understand artifacts, gates, and ownership. - **Quick Reference** (`docs/sdd-add-field-guide.md`): Choose commands by scenario. - **Scenario Playbook** (`docs/sdd-add-scenario-playbook.md`): Handle greenfield, brownfield, errors, and handoff. - **Architecture Profile Guide** (`docs/architecture-profile-guide.md`): Select bindings and technical commands. - **Execution Guide** (`docs/multi-agent-orchestration-guide.md`): Execute tasks/features and worker orchestration. ## Canonical Sources - `.claude/skills/`: Contracts for commands. - `.claude/skills/_shared/ai-review-protocol.md`: Review protocol, methodology, and Action Record. - `.sdd/architecture-profile.md`: Tech bindings and exact verification commands. - `CONSTITUTION.md`: Hard rules on security, architecture, and quality. - `AGENTS.md`: Agent scope and permissions. ## Utility Scripts - `scripts/self-heal.sh`: Runs an approved command to collect evidence; does not modify source or retry. - `scripts/template-smoke.sh` / `.ps1`: Static checks for files, links, and policy contracts. - `scripts/adopt.*`: Adopt the template into an existing repository. - `scripts/update.*`: Pull updates from the template; governance files are staged for human review.