AgentSkillsCN

r-package-skill-creator

通过搭建新的技能文件夹，并可选地从已安装的R包中提取轻量级参考转储（导出内容、帮助索引、小册子），为特定的R包/库（如lme4、dplyr、ggplot2）创建Codex技能。当用户提出“为R包X制作技能”、“为R库搭建Codex技能”，或希望为R包技能统一制定可重复使用的模板时使用。

SKILL.md
--- frontmatter
name: r-package-skill-creator
description: Create Codex skills for specific R packages/libraries (e.g., lme4, dplyr, ggplot2) by scaffolding a new skill folder and optionally extracting lightweight reference dumps (exports, help index, vignettes) from the installed package. Use when asked to “make a skill for R package X”, “scaffold a Codex skill for an R library”, or to standardize a repeatable template for R-package skills.

R Package Skill Creator

Quick start

  1. •Pick a package name (e.g., lme4) and an output folder (e.g., dev/codex/skills/).
  2. •Run python scripts/scaffold_r_pkg_skill.py lme4 --out dev/codex/skills.
  3. •Edit the generated skill’s SKILL.md to add:
    • •3–6 “common workflows” users actually ask for
    • •3–6 troubleshooting bullets for common errors
  4. •Validate the generated skill folder (see “Validation” below).

Workflow

1) Scaffold the skill

Use the bundled script to generate a consistent skeleton (SKILL.md, agents/openai.yaml, and references/ dumps when the package is installed).

Recommended naming convention for generated skills:

  • •r-<pkg> (example: r-lme4)

2) Draft the trigger description

In the generated skill’s YAML frontmatter description, include:

  • •The package name, and the actual tasks users ask for (not package history)
  • •A few key function names (when useful)
  • •Typical contexts: debugging .R / .Rmd / Quarto code, interpreting output, choosing options

Keep it short but trigger-friendly; the description is the main trigger.

3) Curate the generated SKILL.md

Use the skeleton as a starting point, then add:

  • •Common workflows: 3–6 canonical tasks, each with a minimal code snippet (≤ ~10 lines)
  • •Troubleshooting: common errors/warnings and what to check first
  • •Decision points: “If you need X, use function Y; otherwise use Z”

Do not paste whole help pages into SKILL.md; put bulk material under references/ and link to it.

4) Add references (progressive disclosure)

The scaffold script can write:

  • •references/package-description.txt
  • •references/exports.txt
  • •references/help-index.txt
  • •references/vignettes.tsv (when present)

Keep these as “raw dumps” to consult when writing the curated sections.

5) Validation

Run the system validator if available in your environment:

python $CODEX_HOME/skills/.system/skill-creator/scripts/quick_validate.py <path/to/generated-skill>

If $CODEX_HOME isn’t set, run the validator you have access to (or just ensure the YAML frontmatter has only name and description).

Script

scripts/scaffold_r_pkg_skill.py

Create a new r-<pkg> skill folder.

Example:

python scripts/scaffold_r_pkg_skill.py lme4 --out dev/codex/skills

Options:

  • •--skill-name r-lme4 to override the default r-<pkg> naming
  • •--force to overwrite an existing skill folder
  • •--no-references to skip extracting package docs (still scaffolds the skill)