AgentSkillsCN

experiment-provenance

通过可复现的运行元数据、产物指针以及决策日志,记录实验的来源,为科学论断提供可靠依据。

SKILL.md
--- frontmatter
name: experiment-provenance
description: Capture experiment provenance with reproducible run metadata, artifact pointers, and decision logs for scientific claims.

Experiment Provenance

Use this skill when a task produces evidence that may become a manuscript claim.

When to use

  • "track provenance"
  • "log this experiment"
  • "record reproducible run details"
  • "capture evidence for manuscript claim"

Required provenance fields

  1. Objective and hypothesis.
  2. Exact command(s) executed.
  3. Input files, config, and environment assumptions.
  4. Output artifact paths.
  5. Validation status (tests/build/render).
  6. Claim boundary and uncertainty notes.

Workflow

  1. Record run intent before execution.
  2. Execute with deterministic commands where possible.
  3. Store output artifact paths, not just summaries.
  4. Link provenance notes to plans/main-plan-activity.md and plans/main-plan-log.csv.
  5. Mark evidence quality: strong, partial, or insufficient.