Main
- Use the sequential spec flow the README describes (Requirements → Design → Tasks) to keep agent work bounded and reviewable.
- Treat approvals as a gate: request review via the dashboard, capture revisions, then let the agent implement the next approved task only.
- Set a dedicated home directory when needed (README shows
SPEC_WORKFLOW_HOME=...) so workflow state stays in-repo and auditable.
Source-backed notes
- README positions it as an MCP server for structured spec-driven development with a real-time dashboard and a VS Code extension.
- README lists key features like approval workflow, task progress tracking, and implementation logs with code statistics.
- README lists multi-language support and explicitly states it’s available in 11 languages.
FAQ
- Do I need the VS Code extension?: No—the MCP server + web dashboard work without it; the extension is an optional UX.
- Where is the workflow data stored?: Configure a home directory (e.g.,
SPEC_WORKFLOW_HOME) so state lives where you expect. - What’s the minimal way to start?: Run
npx -y ... /path/to/projectas an MCP server and open the dashboard to create your first spec.