---

name: remote-and-scripts description: "Routes Mars remote.spawn, ExecutableTuple, fetch_log, and script-run workflow requests to the verified remote-execution guidance." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Remote Execution and Script Workflows

Use this sub-skill when the user wants Mars remote functions, nested DAGs, returned logs, or a script-run style workflow. It is the right place for mars.remote.spawn, ExecutableTuple, fetch_log, and the run_script contract.

Trigger phrases

• "Run these functions in parallel with Mars."
• "How do I use mars.remote.spawn?"
• "How do I fetch logs from a Mars task?"
• "How do I run a Python script on Mars workers?"
• "How do I pass one spawned result into another?"

What belongs here

• mars.remote.spawn and nested dependency graphs.
• mars.remote.ExecutableTuple fan-in.
• fetch_log usage and log offsets.
• run_script and script-style workflow guidance.

What stays elsewhere

• Tensor/DataFrame execution -> tensor-dataframe-core.
• Learner/integration APIs -> learn-and-integrations.
• Ray/GPU/Kubernetes/YARN or CLI help -> deployment-and-backends.

Read these bundled files

• references/api-reference.md for the verified remote signatures.
• references/workflows.md for fan-out/fan-in and log retrieval patterns.
• references/troubleshooting.md for the common execution and log failures.
• scripts/check_mars_remote.py for a safe local smoke.

Minimal route

1. Import mars.remote as mr.
2. Wrap a function with mr.spawn(...).
3. Call .execute().fetch() when you need the result.
4. Use mr.ExecutableTuple([...]).execute().fetch() for explicit fan-in.
5. Use fetch_log() only when the session is distributed and logs are expected to come back to the client.

Common decisions

• Keep the function body and its arguments tiny when demonstrating the API.
• Use nested spawn only when one remote function depends on another remote result.
• Treat run_script as a workflow contract, not as a reason to depend on the original repo's sample scripts.
• If the user asks for a quick check rather than a full cluster, prefer the bundled smoke helper and a local spawn example.

Quality bar

A future agent should be able to explain how remote tasks are composed, executed, and retrieved; how to fan out and fan in results; how logs are retrieved; and when the script-run contract is appropriate.