metaskills

Portable, installable skills for building better AI agents, workflows, and agent capabilities directly inside your own repositories.

metaskills is for adopters who want practical, reusable agent-building workflows without adopting a new platform. Install a skill, use it in your project, and keep the generated assets in your own repo. MetaSkills 1.1 expands agent-builder with an agent maturity model, harness adapters, handoff contracts, subprocess/job templates, continuous improvement artifacts, and MCP/OpenAPI safety guidance.

Website: https://movetheneedle.info/metaskills/ · Source: https://github.com/smota/metaskills

AI summary

MetaSkills is an open-source toolkit of installable skills that help developers build better AI agents, reusable workflows, and agent capabilities directly inside their own repositories. It helps agents grow from prompt-only helpers to knowledge, tool-enabled, action, orchestrating, and learning agents while keeping a portable file-based core. It is useful for OpenAI/Codex-style agents, Claude, GitHub Copilot, Gemini, Pi, Microsoft Foundry, and other tools that can read local instruction files. It is not a hosted runtime or agent management platform.

Who it is for

• AI engineers building custom agents or agent workflows.
• Open-source maintainers documenting reusable project practices.
• Developer tooling builders packaging repeatable agent capabilities.
• Prompt and agent framework creators who want inspectable examples.
• Teams adopting OpenAI/Codex, Claude, GitHub Copilot, Gemini, or similar coding agents.

Why this matters

Agent-building knowledge is often trapped in one-off prompts, private chats, or an individual developer's habits. metaskills turns that knowledge into portable, inspectable, version-controlled files: skills, templates, checklists, examples, and agent definitions that teams can review, adapt, and improve in Git.

Quick start

List the skills available from this repository:

npx skills add https://github.com/smota/metaskills --list --full-depth

Install one or more skills:

npx skills add https://github.com/smota/metaskills --skill agent-builder --full-depth
npx skills add https://github.com/smota/metaskills --skill skill-creator --full-depth

Install without prompts by passing a target agent, for example:

npx skills add https://github.com/smota/metaskills --skill agent-builder --agent claude-code --full-depth -y

Each installed skill is self-contained and includes its own SKILL.md, README, templates, checklists, references, and examples. The exact install path depends on the selected target agent.

Skill catalog

Skill	Use when	Produces	Install command
agent-builder	You want to create an agent, copilot, coding assistant, tool-enabled assistant, knowledge-focused assistant, or orchestrating agent for another project.	agents/<agent-name>/README.md, AGENT.md, maturity scorecard, runtime capabilities, tools/actions, handoff/job contracts, adapters, evals.	npx skills add https://github.com/smota/metaskills --skill agent-builder --full-depth
skill-creator	You want to turn a workflow, prompt, document, transcript, script, or existing sub-agent into an installable skill.	skills/<skill-name>/SKILL.md, README, templates, checklists, references, examples.	npx skills add https://github.com/smota/metaskills --skill skill-creator --full-depth

Examples

• Build a release notes agent
• Create a support triage skill
• Build a database migration review agent
• Build a repository onboarding agent

See the examples index for a quick comparison.

Available skills

agent-builder

Build portable agents from a plain-language goal.

It helps:

• turn a natural-language idea into an agent specification
• choose a maturity level from Level 0 prompt agent to Level 5 learning agent
• define name, description, instructions, and boundaries
• identify knowledge sources
• design tools/actions with MCP/OpenAPI safety contracts
• document runtime capabilities, harness adapters, and fallback behavior
• model agent-to-agent handoffs and subprocess/job execution when needed
• create starter prompts, evals, feedback loops, benchmarks, and acceptance criteria
• package the result for downstream projects

Try prompts like:

Use the agent-builder skill to design an agent that drafts release notes from GitHub issues and changelog fragments.

Create a knowledge-only onboarding agent for this repository.

Design a tool-enabled agent that reviews database migrations before release.

skill-creator

Create, evaluate, improve, benchmark, and evolve reusable skills.

It helps:

• create production-ready skill packages
• keep skills efficient with progressive disclosure
• generate eval specs and pass/fail cases
• plan improvements from feedback or failed evals
• benchmark versions or alternatives
• support optional learning logs
• support optional privacy-safe telemetry plans

Try prompts like:

Use the skill-creator skill to turn this support triage workflow into a reusable skill.

Evaluate this existing skill and suggest improvements with acceptance criteria.

Create a benchmark plan comparing v1 and v2 of this skill.

Use in a downstream project

1. Open the project where you want the skill available.
2. Run the npx skills add install command from that project root.
3. Ask your coding agent to use the installed skill by name.
4. Review the generated files.
5. Keep generated agents or skills in your project repo.
6. Re-run the install command when you want the latest version.

Interoperability and trust

metaskills is file-based and vendor-neutral. It can support OpenAI/Codex-style coding agents, Claude, GitHub Copilot, Gemini, and other agent harnesses that can read project instructions. See Interoperability and trust model.

What this repo is not

metaskills is not a hosted AI platform, agent runtime, management system, deployment platform, issue tracker for downstream projects, or central control plane. It is a small collection of reusable skills and references that help other projects build their own agents and skills.

Contributing

Contributions are welcome when they improve reusable skills, agents, toolsets, examples, components, or docs.

Start with:

• Contributing guide
• Create a skill
• Review a skill
• Release process
• GitHub discoverability checklist

Repository structure

skills/      Installable skill packages.
agents/      Native agent definitions and examples, when needed.
toolsets/    Tool integration guides, schemas, wrappers, and examples, when needed.
components/  External repositories linked as Git submodules plus .metaskills.md metadata.
examples/    Cross-skill examples that are not owned by one skill.
docs/        Repository architecture, conventions, and workflow runbooks.

More documentation

• Documentation index — adopter, builder, and contributor navigation.
• Repository architecture — structure, asset ownership, and implementation details.
• Component submodule workflow — how external inspiration/reference repos are linked.
• Interoperability and trust model — compatibility, boundaries, and safety model.

Inspirations and references

These skills are inspired by and reference the following projects and capabilities:

• Microsoft 365 Copilot Agent Builder — natural-language agent creation, configure/review workflow, knowledge sources, actions, and starter prompts.
• FrancyJGLisboa/agent-skill-creator — plain-English workflow-to-skill strategy, validation, security scanning, evals, and cross-platform skill packaging.
• Claude Skill Creator — Create, Eval, Improve, and Benchmark lifecycle for skill development.
• metaskills/skill-builder — production-ready skill structure, progressive disclosure, and sub-agent-to-skill conversion guidance.