execution-ml-executor

name	execution/ml-executor
description	Internal: ML Executor. Implements ML models, training pipelines, inference code, data processing. Read spec and segment. Write evidence on completion.

dynos-work: execution/ml-executor

Spawn the ml-executor agent with the user's prompt as the instruction.

Ruthlessness Standard

• Do not let the executor hide weak assumptions behind model language or data-science jargon.
• If data shape, training flow, or inference contracts are unclear, the agent must inspect them before editing.
• Do not accept a change that improves one metric while making reproducibility, safety, or serving behavior ambiguous.

What to pass

Pass the user's full prompt verbatim. Do not summarize or sanitize it. Prepend a short hard wrapper that tells the agent to verify dataset/model contracts, fix the real mechanism, and only claim success with concrete evidence.