Create New Skills (Skill-Authoring Meta-Skill)
A meta-skill that teaches Claude Code how to author new Agent Skills, turning skill creation into a repeatable, best-practice workflow.
A meta-skill for creating new skills, requiring reading official docs on the three-level progressive-disclosure structure before starting — pure documentation generation.
What does this skill do, and when should you use it?
This is one of four skills bundled in the disler/claude-code-hooks-multi-agent-observability repository, located at .claude/skills/meta-skill/. It does not monitor agents; instead it is a guide for having Claude write skills: it instructs Claude to first read three documentation files in the bundled docs/ directory, then follow an eight-step workflow (define purpose, create directory, write SKILL.md, write instructions, write examples, add supporting files, test, commit to git). It covers the three-level progressive disclosure architecture, naming conventions, frontmatter limits (name ≤64 chars, description ≤1024 chars), the optional allowed-tools field, and four complete worked examples. It suits teams that want to package domain expertise into reusable, git-shared Claude Code skills.
Reads three required docs from the bundled docs/ directory (agent skills guide, architecture overview, design-principles blog post); asks the user four clarifying questions to scope the new skill; creates a directory under .claude/skills/<skill-name>/ or ~/.claude/skills/; generates a SKILL.md with YAML frontmatter (name, description, optional allowed-tools) structured as Prerequisites, Workflow, and Examples; optionally creates supporting files under scripts/, templates/, or guidelines/ and references them by relative path; validates the structure with ls and head commands; and finally commits the skill via git so team members receive it automatically on pull.
- An engineering team wants to package an internal code review process (e.g., PEP 8 checks) as a skill shared via git
- A data analyst wants to bundle a CSV analysis workflow with pandas/matplotlib scripts into a reusable skill
- A technical writer wants a multi-file documentation skill driven by a company style guide, using progressive disclosure to manage context
- A maintainer of an existing skill wants to extend it, such as adding a spell-correction mapping file to a transcription skill
- A developer new to the Agent Skills spec wants a structured introduction to SKILL.md conventions
What are this skill's strengths and limitations?
- Complete process: an eight-step workflow from requirement clarification to git commit, with concrete commands and templates at every step
- Ships with three required-reading docs explaining progressive disclosure, so no external material is needed
- Four end-to-end examples (simple review, script-based analysis, multi-file docs, skill extension) whose structure can be copied directly
- Distills patterns and anti-patterns from real skills, including good/bad description writing
- Pure Markdown guidance with no script dependencies, usable in any Agent Skills-compatible client
- Strongly Claude Code-oriented: .claude/skills/ placement and git-based team sharing may not hold on other platforms
- The parent repo is a multi-agent observability system; this meta-skill is incidental, and the repo's main documentation (Bun/Vue dashboard, hook events) does not help you use it
- No dedicated test suite or quality validation — output quality depends on Claude's execution
- License is listed as unknown; verify before commercial use
- Script dependencies shown in examples (pandas, matplotlib, etc.) are illustrative; the skill itself ships none of them
How do you install this skill?
The skill is part of the repository's .claude/ directory. Per the README, copy the entire .claude directory to your project root: cp -R .claude /path/to/your/project/; the skill then lives at .claude/skills/meta-skill/. The repo documents no standalone installation for this single skill; to take only it, manually copy the meta-skill folder (SKILL.md plus docs/) into your project's .claude/skills/. License is listed as unknown.
How do you use this skill?
Trigger it in Claude Code — no CLI steps required. Example prompts: "Create a skill that reviews Python code for best practices" or "Turn our style guide into a documentation skill." The SKILL.md description states it should trigger when the user wants to create a new skill, extend Claude's capabilities, or package domain expertise. Once triggered, it directs Claude to read the docs/ directory and then walk you through definition, creation, testing, and git commit.
How does this skill compare with similar options?
Compared with Anthropic's official Agent Skills documentation, this skill compresses the spec into a step-by-step guide with examples suited to in-conversation execution; however, its Claude Code directory conventions and git sharing mechanism make it less portable than the plain spec.