Give your agent focused workflows, domain knowledge, and reliable tools without rewriting the same instructions every time.
A skill is a self-contained package that teaches an AI agent how to perform a specific task. It can combine detailed instructions with scripts, reference material, configuration, tests, and reusable assets.
Each skill starts with a SKILL.md file that tells an agent:
- when to use the skill;
- which workflow to follow;
- what inputs and tools it needs;
- what a successful result looks like.
Tip
Think of a skill as an installable playbook: focused enough for one job, complete enough to reuse.
Installing a skill can be as simple as sharing its URL with an agent that supports skills.
Open a skill from the library below and copy the URL from your browser.
Install this skill: https://github.com/omnedia/skills/tree/master/crm/research-qualified-leads
Describe your task normally, or mention the skill by name:
Use the research-qualified-leads skill to find 20 qualified leads.
Important
Install the complete skill directory, not only SKILL.md. Supporting scripts, references, configuration, or assets may be required for the skill to work.
| Skill | What it does |
|---|---|
| Research Qualified Leads | Researches, qualifies, contact-enriches, deduplicates, and exports evidence-backed sales leads. |
| Skill | What it does |
|---|---|
| Omnedia Captions | Creates animated, transparent caption overlays in editable Remotion projects from timed transcripts or text with matching audio. Requires the Remotion plugin in Codex. |
| Omnedia Motion Graphics | Creates sparse editorial collage overlays, layered parallax, paper treatments, callouts and diagrams as silent alpha clips with editable Remotion projects. Requires the Remotion plugin in Codex. |
More skills and categories are on the way.
Agent support and accepted URL formats can differ. If direct installation does not work, clone the repository:
git clone https://github.com/omnedia/skills.gitThen copy the complete directory of the skill you want into your agent's local skills folder. Restart or reload the agent if it does not discover the new skill automatically.
Some skills require extra inputs, tools, credentials, or dependencies. Check the skill's SKILL.md for its exact requirements.
skills/
|-- <category>/
| `-- <skill-name>/
| |-- SKILL.md # Instructions and metadata
| |-- skill.yaml # Authoritative version and canonical location
| |-- README.md # Usage and version bump guidance
| |-- scripts/ # Optional executable helpers
| |-- references/ # Optional supporting documentation
| |-- assets/ # Optional templates and resources
| `-- tests/ # Optional automated tests
`-- README.md
Every skill is self-contained, so its directory can be installed independently.
Skills in this repository are versioned independently using Semantic Versioning. Every skill has its own skill.yaml containing its name, version, canonical repository (https://github.com/omnedia/skills), and repository-relative path. The version in that file is authoritative for that skill; there is no replacement global repository version.
All skills existing when this convention was introduced start at 1.0.0. Use PATCH for fixes, clarifications and prompt refinements, MINOR for backward-compatible new capabilities, and MAJOR for breaking changes to behavior, interfaces, workflows, inputs, outputs or compatibility. When a change warrants a release, bump that skill's skill.yaml in the same change. Read the current version from skill.yaml rather than duplicating it in documentation. Updating one skill does not require bumping unrelated skills; leave their versions unchanged.
Each skill checks its installed version against its canonical metadata on master when access is available, at most once per conversation/session unless explicitly requested again. Newer versions are reported with both version numbers. Updates require explicit user approval and are never installed automatically. A failed or unavailable check does not block normal operation or justify claiming the skill is current.
New skills and improvements are welcome. A good contribution should:
- live in an appropriate category;
- use a clear, kebab-case directory name;
- include a
SKILL.mdwithnameanddescriptionYAML frontmatter; - document prerequisites, inputs, outputs, and safety constraints;
- keep required scripts, references, and assets inside the skill directory;
- include tests for deterministic scripts where practical;
- add the new skill to the library in this README.
Skills can instruct agents to execute commands, access services, or modify files. Review third-party instructions and scripts before installing them, and only run code you trust.
Distributed under the MIT License. You are free to use, modify, and distribute these skills in accordance with its terms.
Build better agents, one skill at a time.