An agentic planning primitive with a plugin ecosystem — not a CorridorStudies replacement.
Each skill is a directory with skill.yaml + skill.py. See CONTRIBUTING.md — 4 steps. Schema is enforced by the registry (name pattern, kind enum, JSON-Schema inputs, Skill subclass, import isolation).
POST /api/sessionPOST /api/{sid}/runGET /api/skills and GET /api/skills/{name}/schemaPOST /api/{sid}/skills/{name}:run — headless, no agentGET /ws/{sid} — JSON events (tool_call, tool_result, layer, violation, iteration, queue, done, error)Our UI is one client; your agent can be another.