Universal AI Skill Library

One searchable, bilingual library for discovering and installing reusable Agent Skills across Claude Code, OpenAI Codex, and OpenCode.

An Agent Skill is a folder centered on a SKILL.md file. It teaches an AI coding agent when and how to perform a repeatable workflow—for example code review, research, testing, data analysis, security checks, design, documentation, or automation. A skill may also include references, templates, scripts, or platform-specific metadata.

This repository gives beginners clear download paths and gives advanced users transparent compatibility, risk, provenance, and package information.

These are catalog mappings, not a promise that every dependency or tool was executed successfully on every platform. Read the platform note and inspect a skill before granting permissions.

Finance and investing: the validated 17-skill subset also has a focused repository: Universal AI Finance Skills. It uses calibrated scenarios and risk gates; it does not promise returns.

Start here

Download for your agent

The recommended bundles already contain the correct hidden directory tree. Download the package matching the agent you use, inspect it, and extract it into your home folder or project root.

New to GitHub? Use the blue download links above. The green Code → Download ZIP button downloads the repository source; it is not the ready-to-install platform bundle.

How it works

1. Discover: search 588 names by task or category.
2. Compare: check the platform note, requested tools, network/API use, destructive patterns, and license signal.
3. Download: choose one skill ZIP, a curated platform bundle, or an expanded archive.
4. Install and verify: extract to the correct root, confirm the SKILL.md layout, and compare the SHA-256 checksum.

Quick installation

1. Choose personal or project scope

OpenCode can also discover compatible skills in Claude and Codex skill roots. The OpenCode bundle uses its native .opencode/skills layout so each package stays explicit and predictable.

2. Check the final folder shape

Every installed skill must have SKILL.md directly inside its own folder:

<skills-root>/
└── example-skill/
    ├── SKILL.md
    ├── references/     optional
    ├── scripts/        optional
    └── assets/         optional

Avoid accidental double nesting such as example-skill/example-skill/SKILL.md.

3. Reload and test

Most hosts discover valid skills automatically. If a new skill does not appear, verify the path and YAML frontmatter, then restart the host. Ask naturally or name the skill directly, for example:

Use code-review-and-quality to review this change and prioritize actionable defects.

See the complete installation guide for individual downloads, updates, uninstalling, checksum verification, and troubleshooting.

Browse by category

Browse through the interactive English/Turkish catalog, the complete English table, or the Turkish table. Every catalog row includes a description, platform note, capability/risk signals, license status, and direct downloads.

Curated vs expanded

Curated bundles — recommended

• Keep the initial top-level skill list manageable.
• Include 151 Claude, 151 Codex, or 154 OpenCode skills.
• Include skill-library-router, which exposes the embedded catalog progressively instead of flooding the agent's discovery metadata.
• Best for normal daily use.

Expanded bundles — advanced/offline

• Install every compatible catalog name directly: 585 / 585 / 588.
• Useful for offline archives, auditing, or custom packaging.
• Can crowd a host's skill-discovery context and make selection noisier.

Preserved variants

The library keeps 21 conflicting embedded variants under skills/archive-variants/ instead of silently merging or overwriting them. These are reviewable archive records, not automatically selected replacements.

Safety, provenance, and licenses

Skills are instructions and may reference scripts, package installation, external APIs, credentials, privileged operations, or destructive commands. Before use:

• Read SKILL.md and any scripts or referenced files.
• Review required tools, dependencies, network access, and external services.
• Keep secrets, private code, and sensitive data out of untrusted services.
• Require explicit confirmation before installation, deletion, publication, purchases, or privileged changes.
• Test high-impact skills in an isolated project or disposable environment.

License scope matters: 405 catalog entries have no local redistribution-license signal. The repository's LICENSE.md covers repository-authored catalog metadata, documentation, website, and packaging code—not all third-party skill bodies. Each skill retains its own upstream terms. Read THIRD_PARTY_NOTICES.md and PUBLISHING.md before reuse or redistribution.

To verify a downloaded file on Windows:

Get-FileHash .\universal-ai-skill-library-codex.zip -Algorithm SHA256

On macOS/Linux:

sha256sum universal-ai-skill-library-codex.zip

Compare the result with manifests/SHA256SUMS.txt.

Repository architecture

skills/common/                 portable common core
skills/platforms/codex/        Codex metadata overlays or Codex-only skills
skills/platforms/opencode/     OpenCode-only skills
skills/archive-variants/       preserved conflicting embedded versions
packages/common/               individual portable ZIP files
packages/<platform>/           platform-specific individual ZIP files
release-assets/                curated and expanded release bundles
manifests/                     JSON, CSV, checksums, and repository metadata
docs/                          bilingual searchable static catalog

The source catalog's hosts field is treated as a mapping signal—not runtime proof. Platform-specific behavior is kept visible rather than silently flattened into one file.

Documentation map

FAQ

No. There are 588 unique names; 585 map to Claude Code, 585 to Codex, and 588 to OpenCode. Format compatibility does not prove that every dependency, external service, or platform-specific behavior was executed successfully. Read the catalog note for the skill.

Start with the curated bundle. Use expanded only when you intentionally need a complete offline archive or plan to audit/repackage the collection.

The most common causes are the wrong root folder, double nesting, a lowercase or missing SKILL.md, invalid YAML frontmatter, or a host that needs restarting after its skill root was first created. Follow INSTALL.md.

No. The repository-authored catalog and packaging layer use the scope described in LICENSE.md. Third-party skill bodies keep their own terms, and an unspecified signal is not an open-source grant.

Contributing and support

• Improvements and provenance corrections: open an issue or read CONTRIBUTING.md.
• Security concerns: follow SECURITY.md; do not publish sensitive exploit details in a public issue.
• Rights or takedown requests: provide the affected path, ownership evidence, and requested action.

If this library saves you time, star the repository so more builders can discover it.