Claude-skill-registry writing-skills

Use when troubleshooting skill issues, understanding CC platform limitations, or reviewing skills for CC-specific gotchas. For skill creation tutorials, use plugin-dev:skill-development.

install
source · Clone the upstream repo
git clone https://github.com/majiayu000/claude-skill-registry
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/cc-writing-skills" ~/.claude/skills/majiayu000-claude-skill-registry-writing-skills-21bf27 && rm -rf "$T"
manifest: skills/data/cc-writing-skills/SKILL.md
source content

Writing Skills

Scope

This skill covers Claude Code platform gotchas and limitations. For different purposes, use:

  • Comprehensive tutorials:
    plugin-dev:skill-development
  • TDD methodology:
    superpowers:writing-skills
  • This skill: Bug workarounds, version-specific behavior, issue tracking

Facts

  • The frontmatter
    name
    is what appears in autocomplete list when typing
    /
    in session. So
    name: foo
    would allow user to submit
    /foo
    to trigger that skill.
  • The skill directory name appears printed in session when skill is activated by the agent. E.g. dir
    /foobar
    would yield e.g.
    ⏺ Skill(foobar)
    ⎿  Initializing…
    
  • Skill(foobar)
    is also the syntax for granting permission to the skill in config
  • Development Related
    • Skills created or modified in ~/.claude/skills or .claude/skills are immediately available without restarting the session (added in v2.1.0)

Key Rules (from Anthropic guide)

Frontmatter:

  • name
    : Max 64 chars, lowercase letters/numbers/hyphens only, no reserved words ("anthropic", "claude")
  • description
    : Max 1024 chars, non-empty, third person, critical for discovery
    • Good: "Processes Excel files and generates reports"
    • Avoid: "I can help you process Excel files"
  • Use gerund form (verb + -ing) for names:
    processing-pdfs
    ,
    analyzing-spreadsheets

Body:

  • Keep SKILL.md under 500 lines
  • "Claude is already very smart. Only add context Claude doesn't already have."
  • Paths are relative to skill directory — use
    scripts/foo
    not
    .claude/skills/my-skill/scripts/foo

File Structure

my-skill/
├── SKILL.md          # Main (<500 lines)
├── reference/        # Detailed docs (loaded on demand)
├── scripts/          # Executable (zero-token)
└── assets/           # Static files (banners, templates)

Naming: Directory name = skill name (kebab-case). Opening

# Title
= titleized skill name.

Skill Body Headings

HeadingPurpose
## CRITICAL
Safety rules at top
## Use Cases
Refine when skill applies
## Requirements
Deps, env vars, how to handle absence
## Steps
Sequential workflow
## Operations
Menu of actions
## Examples
User prompt → Claude action
## Reference
Internal docs
## Notes
Gotchas, tips
## Appendix
Implementation details

Tools for Skills

ToolPurposeReference
AskUserQuestionInteractive prompts (max 4 options)tool-ask-user-question.md
TodoWriteTask checkboxestool-todo-write.md
BashExecute bundled scripts (zero-token)pattern-script-bundling.md
ReadLoad reference files on demandpattern-progressive-disclosure.md

Frontmatter Options

FieldPurposeReference
name
,
description
Identity and discoveryconfig-frontmatter.md
allowed-tools
Skip permission promptsconfig-allowed-tools.md
context: fork
Run in isolated subagentconfig-context-fork.md
hooks
PreToolUse/PostToolUse/Stopconfig-hooks.md
user-invocable
Show in slash menuconfig-frontmatter.md
disable-model-invocation
User-triggered onlyconfig-frontmatter.md

Platform Limitations

Cannot control:

Patterns

PatternReference
Code examplespattern-code-examples.md
Progressive disclosurepattern-progressive-disclosure.md
Script bundlingpattern-script-bundling.md

When Reviewing Skills

Apply criteria from both this skill (CC platform) and

superpowers:writing-skills
(methodology).

Platform checklist:

  • Frontmatter valid? (name format, description length, third person)
  • SKILL.md under 500 lines?
  • Reference files used for detailed content?
  • Code examples have language tags?
  • Known limitations acknowledged? (output collapse, status line)
  • Hooks correctly structured if used?
  • allowed-tools
    syntax correct if used?

Methodology checklist (from

superpowers:writing-skills
):

  • CSO applied? (Claude-Specific Optimization)
  • Token-efficient structure?
  • Clear trigger conditions in description?