AgentSkillsCN

scientific-pipeline

设计或重构科学数据管道，构建清晰的文件夹结构、可复现的环境、编排工具（Snakemake/Make），并生成诊断图表。

SKILL.md
--- frontmatter
name: scientific-pipeline
description: Design or refactor a scientific data pipeline with clear folder structure, reproducible environments, orchestration (Snakemake/Make), and diagnostic plots.

Objectives

  • •Separate data processing from visualization.
  • •Make the pipeline deterministic, re-runnable, and inspectable.
  • •Make it obvious where data came from and whether outputs are stale.

Checklist

  1. •Directory structure
    • •Ensure data/raw, data/processed, data/generated, src, tests, scripts, reports exist.
  2. •Environment
    • •Ensure dependencies are pinned/recorded (uv/conda/docker).
  3. •Orchestration
    • •Prefer Snakemake or Make (or, at minimum, one scripts/run_pipeline.sh entrypoint).
    • •Define inputs/outputs per step.
  4. •Validation
    • •Add fast sanity checks and cheap diagnostics (saved PNGs).
    • •Promote critical checks to tests.
  5. •Reproducibility
    • •No hard-coded paths.
    • •Fixed random seeds where randomness is used.

Deliverables

  • •A single documented entrypoint to run the pipeline.
  • •Tests for at least the core transformation.
  • •A short note in docs/DECISIONS.md for any scientific choice.