<2-3 sentences summarising what the project addresses and solves, the target audience, and the benefit.> Addresses AI coding assistants' limitations (context overload, lack of structure) with a Spec-Driven Development (SDD) orchestrator and 9 specialized sub-agents. Targets engineers and power users, enabling structured feature development, reducing AI hallucinations, and ensuring code quality through a defined workflow.

How It Works

A lightweight orchestrator delegates tasks to specialized, fresh-context sub-agents, preventing context compression and hallucinations. Development follows a Directed Acyclic Graph (DAG) for phases like exploration, specification, design, implementation, and verification. Artifacts (specs, designs) are managed via a pluggable persistence layer: engram (default), openspec (file-based), or none (ephemeral).

Quick Start & Requirements

Install via ./scripts/install.sh, which copies Markdown "skill" files for AI assistants. Zero dependencies. Supports Claude Code, OpenCode, Gemini CLI, Codex, VS Code (Copilot), Antigravity, and Cursor. Setup involves integrating skills and orchestrator instructions into the chosen AI tool's configuration.

Highlighted Details

• 9 Specialized Sub-Agents: Dedicated skills for each SDD phase (Init, Explorer, Proposer, Spec Writer, Designer, Task Planner, Implementer, Verifier, Archiver).
• Pluggable Artifact Store: Flexible persistence options: engram (recommended), openspec, or none.
• Delta Specs: Changes are defined as deltas (ADDED/MODIFIED/REMOVED) that merge into main specs upon archiving, providing a clear audit trail.
• v2.0 Upgrades: sdd-apply gains TDD workflow support; sdd-verify adds real test execution and a spec compliance matrix.

Maintenance & Community

Contributions via Pull Requests are welcomed. The project is maintained by Gentleman Programming. No specific community channels or contributor lists are detailed in the README.

Licensing & Compatibility

MIT License. Permissive for commercial use and integration with closed-source projects.

Limitations & Caveats

Tools lacking native sub-agent delegation (Gemini CLI, Codex, Cursor) run skills inline, consuming more context. This "Lite" version represents Level 2 of the agent team architecture; Level 3 (Full Agent Teams) with advanced features like shared task queues is not implemented.