Troth

Connects to Claude.ai, Claude Code, Cursor, or any MCP client via SSE at mcp.troth.dev/sse. Breaks high-level specs into discrete tasks with acceptance criteria, bounded file scopes, and verification gates. The agent writing code doesn't verify its own work. Instead, Troth enforces test commands and criteria before marking tasks complete. Generates structured YAML that defines architecture patterns, entry points, dependencies, and which files each task can touch. Use it when you want coding agents to work from a structured map rather than free-form conversation, especially on projects where parallel work needs explicit conflict prevention and every change should trace back to a verified requirement.