TokenRollAI

update-doc

"Use when user says 'update docs', 'sync documentation', 'refresh docs', or after code changes need documenting. Updates llmdoc system based on recent code changes."

TokenRollAI 484 54 Actualitzat fa 9 mesos
GitHub

Instal·la

npx skillscat add tokenrollai/cc-plugin/update-doc

Instal·la des del registre de SkillsCat.

SKILL.md

/update-doc

This skill updates the project's llmdoc documentation to reflect recent code changes.

Pre-fetched Context

  • Llmdoc structure: !find llmdoc -name "*.md" 2>/dev/null | head -50
  • Recent changes (3 commits): !git diff HEAD~3..HEAD --stat 2>/dev/null | head -30
  • Affected files: !git diff HEAD~3..HEAD --name-only 2>/dev/null | head -30
  • Llmdoc index: !cat llmdoc/index.md 2>/dev/null | head -50

Actions

  1. Step 1: Analyze Changes

    • If $ARGUMENTS is provided, use it as the description of what changed.
    • Otherwise, analyze the pre-fetched git diff to understand what changed.
  2. Step 2: Identify Impacted Concepts

    • Map changed files to llmdoc concepts:
      • Config files (.eslintrc, etc.) → reference/coding-conventions.md
      • Auth files → relevant architecture docs
      • New features → may need new docs
    • Create a list of impacted documents.
  3. Step 3: Update Documents

    • For each impacted document, use recorder agent with this prompt:
      Task: Update documentation for <concept_name>.
      Changes: <relevant git diff summary>
      Mode: content-only
      Principle: Use fewest words necessary.
  4. Step 4: Synchronize Index

    • After all updates complete, invoke a single recorder agent to:
      • Re-scan /llmdoc directory
      • Ensure index.md is consistent and up-to-date
      • Mode: full
  5. Step 5: Report

    • Summarize all documents created/updated/deleted.

Update Principles

  • Minimality: Use fewest words necessary
  • Accuracy: Based on actual code, not assumptions
  • No Code Blocks: Reference with path/file.ext:line format
  • LLM-Friendly: Write for machine consumption