Complete AI Training
Sign inGet started

Skill · Business

Rote

Compiles a proven SKILL.md into a deterministic pipeline via the rote CLI, then registers and serves it as MCP tools. Use when the user wants to compile a skill, turn a skill into a pipeline, pick a runtime target, or serve and register compiled pipelines.

Complete AI SkillsLicense: Apache-2.0Added Sep 29, 2026
SalesHuman ResourcesFinance & AccountingProduct Management
Use it in my AI View on TemplatesGrokBot

How to use it

  1. Start your plan and connect your AI once
  2. Ask for the task in your own words, or say it directly:
Use the Rote skill to help me with this.

Without a connection: copy the SKILL.md below into your AI's project instructions.

SKILL.md

Rote

Compiles proven agent skills into deterministic pipelines and serves them as MCP tools. For users who have a working SKILL.md and want it to run without an LLM.

When to use

  • User asks to compile a skill or turn a SKILL.md into a pipeline.
  • User wants to pick a runtime target (dbos, temporal, python, cloudflare, dbos-ts, inngest).
  • User wants to register or serve compiled pipelines as MCP tools.
  • User asks for a pipeline summary or codified fraction after compilation.

Workflows

Identify source skill

Inputs: Path to the skill directory containing SKILL.md, optionally with references/.

  1. Locate the skill directory containing SKILL.md.
  2. Confirm the absolute path with the user before running; compilation costs real time and tokens.
  3. If no SKILL.md exists, stop and ask the user for the correct path.

    4. Check: SKILL.md exists at the confirmed absolute path. Output: Confirmed absolute skill directory path.

Pick runtime target

Inputs: User runtime preference and any existing infrastructure.

  1. Choose a runtime from: dbos (default), temporal, python, cloudflare, dbos-ts, inngest.
  2. Default to dbos when the user has no preference and no existing infrastructure.
  3. State the choice and reason before proceeding.

    4. Check: Runtime stated and accepted by the user. Output: Selected runtime name with rationale.

Run compilation

Inputs: Confirmed skill directory, selected runtime, output directory.

  1. Ensure uv is installed; if not, ask the user to install it through their package manager. Never pipe remote scripts.
  2. Invoke uvx --from 'rote-cli>=0.12.1' rote compile <skill-dir> --runtime <runtime> --out <out-dir>.
  3. Set expectations: a realistic skill takes ~13 minutes and 30-40 agent turns.
  4. Run in background, poll, and surface stderr on failure.

    5. Check: Compilation completes without error; stderr surfaced if it fails. Output: Compiled output directory.

Report pipeline summary

Inputs: compiled/pipeline.yaml and compile-report.md in the output directory.

  1. Read compiled/pipeline.yaml and compile-report.md.
  2. Summarize node-kind counts: pure_function, external_call, llm_judge, agent_loop, hitl_gate.
  3. Report the codified fraction, where files landed, and next steps.
  4. Be honest if the pipeline is mostly agent_loop; do not spin a success story.

    5. Check: Counts and codified fraction match the compiled files. Output: Summary of node-kind counts, codified fraction, file locations, next steps.

Serve and register pipelines

Inputs: Deployed runtime and output directory.

  1. After the runtime is deployed, register with rote register <out-dir> (or with --runtime temporal/cloudflare).
  2. Add the MCP server via claude mcp add --scope user rote -- uvx --from 'rote-cli[serve,dbos]>=0.12.1' rote serve.
  3. Explain that each pipeline becomes <name>, <name>_status, and on DBOS <name>_signal tools.
  4. Note that Claude Desktop snapshots tools at connect time.

    5. Check: Registration succeeds and MCP server is added. Output: Registered pipelines and MCP server configuration.

Tools and data

  • Use Claude Code or Codex CLI (authed) when available.
  • Use uv when available; if not available, ask the user to install it through their package manager.
  • Use rote-cli when available.
  • Use MCP server registration when available.

Guardrails

  • Never write pipeline.yaml or classify nodes; the CLI's compiler agent does that.
  • Never run compilation without confirming the source template path with the user.
  • Never pipe remote install scripts into a shell; ask the user to install uv through their package manager.
  • Do not modify auth environment variables to force API billing; respect the default claude driver unless the user explicitly requests --agent api.
  • Stop when the skill is exploratory or one-off, because there is nothing proven to compile yet.

Getting started

Ask the user for the path to the skill directory containing SKILL.md, and whether they have a runtime preference. Confirm the path and runtime before starting compilation.

Credits

Adapted from an open-source original (Apache-2.0): https://www.aitmpl.com/component/skills/workflow-automation/rote

More business skills

All business skills