AgentSkillsCN

file-safety

制定文件修改规则，防止覆盖重要文件。在覆盖现有文件、删除文件，或向包含重要数据的目录（data/、outs/）写入数据时使用。同样适用于移动或重命名其他脚本所依赖的文件。

SKILL.md
--- frontmatter
name: file-safety
description: File modification rules to prevent overwriting important files. Use when overwriting existing files, deleting files, or writing to directories that contain important data (data/, outs/). Also applies when moving or renaming files that other scripts depend on.
user-invocable: false

File Modification Rules (CRITICAL)

General Rules (All Projects)

Never Overwrite These Files

  1. •Outputs from someone else — files from collaborators, students, or external sources
  2. •Original/source data — anything you did not generate in this session
  3. •Configuration files — unless the user explicitly requests changes

Instead

  • •Create a new file with an updated name (e.g., add _updated, _with_new_column)
  • •Ask the user before modifying if uncertain about the file's origin

Files You CAN Overwrite

  • •Files you created in the current session
  • •Configuration files when the user requests changes

Script Output Integrity

Never create standalone scripts solely to modify outputs from other scripts.

If modifications are needed:

  1. •Modify the original script — if the change belongs upstream, update the source (ask user first)
  2. •Handle in downstream scripts — do the transformation as part of the consuming script's workflow

Do NOT create intermediate "fixer" scripts that modify upstream outputs after the fact — this breaks the reproducibility chain between scripts and their outputs.

When Uncertain

If you're not sure whether a file is safe to overwrite, ask:

  • •"This file exists — was it generated by another script or provided externally?"
  • •"Should I create a new file with a modified name, or is it safe to overwrite?"

Data Science Projects (with data/ and outs/ directories)

These additional rules apply in projects with project-type: data-science or that use the script-organization conventions (numbered scripts, data/ for inputs, outs/ for outputs).

Output Ownership

Each script writes only to its own output folder under outs/. Never write to another script's output folder.

r
# Script 03_plots.qmd writes ONLY to:
out_dir <- here("outs/03_plots")

# NEVER write to another script's folder:
# here("outs/01_analysis/modified_data.rds")  # WRONG

data/ is Read-Only

Scripts never write to data/. This directory contains only external/immutable inputs (raw data, collaborator files, annotations). If your code produced it, it goes in outs/.

Example

r
# BAD: Creating a script to add Gene.short to another script's output
# This disconnects the file from the script that produced it

# GOOD: Downstream plotting script joins against lookup at runtime
limma_res <- readRDS(here("outs/01_analysis/limma_results_annotated.rds")) %>%
  left_join(gene_name_lookup, by = "trinity_gene_id")