What problem does it solve?
Running a proven agent skill repeatedly through an LLM loop is slow, expensive, and non-deterministic. This Skill compiles a SKILL.md-based skill into a deterministic pipeline that runs without an LLM in the loop, then serves it back to Claude as an MCP tool.
Core Features & Use Cases
- Skill Compilation: Runs the rote CLI's LLM compiler agent over a source skill once, emitting a pipeline with typed node kinds (pure_function, external_call, llm_judge, agent_loop, hitl_gate).
- Multi-Runtime Targets: Generates deployable code for DBOS, Temporal, plain Python, Cloudflare Workflows, DBOS TypeScript, or Inngest.
- MCP Serving: Registers compiled pipelines and exposes them via
rote serve as MCP tools with start, status, and HITL-signal operations.
- Use Case: You have a skill you run dozens of times daily. Compile it once with rote, deploy the DBOS runtime, register it, and call it from Claude as a background job that runs for minutes to days without an agent loop.
Quick Start
Compile my skill at ./skills/my-skill into a deterministic DBOS pipeline and register it so I can call it from Claude.