agent-skills-library

Skills Library

Curated bilingual (EN + RU) skills in the universal Agent Skills format

Works with Claude Code, OpenCode, Cursor, Windsurf and every Agent Skills-compatible tool

Live Catalog Site · Releases · Issues · Discussions

---

Install in one line

curl -fsSL https://raw.githubusercontent.com/ssrjkk/agent-skills/main/install.sh | bash

Or clone directly into your agent's skills directory:

What is this?

100 production-grade, bilingual (English + Russian) skills following the universal Agent Skills format — the SKILL.md convention shared by Claude Code, OpenCode, Cursor, Windsurf and other agents. Each skill is a folder with a SKILL.md (primary, English) and an optional SKILL.ru.md (parallel Russian translation). Any agent that implements the Agent Skills spec can consume them directly.

The library ships with a Python SDK + CLI for validation, quality scoring, cataloging and search, all enforced in CI.

Quick Start

# Install the SDK + CLI from PyPI
pip install agent-skills-library

# Explore the library (both `agent-skills` and `claude-skills` work)
agent-skills stats          # Library statistics
agent-skills search <query> # Search skills
agent-skills validate       # Validate all skills (EN + RU)
agent-skills quality        # Quality analysis
agent-skills catalog        # Rebuild skills_catalog.json

Using the skills with your agent

Every skill is a standard Agent Skills directory:

.claude/skills/
  {domain}/
    {skill-name}/
      SKILL.md        <- English (primary)
      SKILL.ru.md     <- Russian (parallel)

To install a skill into another agent, point it at the same folder — the file format is identical, only the root directory changes:

Frontmatter is strictly validated for cross-agent portability in CI (scripts/check_agent_interop.py): portable name, single-line description, no duplicate names.

Stats

Only skills meeting the quality bar (Grade A, no validation errors) are kept in main. Everything else is archived under the v1.0-legacy tag.

Skills by domain

SDK

Python

from claude_skills.catalog import CatalogBuilder
from claude_skills.validator import ValidationPipeline
from claude_skills.quality import QualityAnalyzer

catalog = CatalogBuilder().build_catalog()
print(f"{catalog.metadata.total_skills} skills, {catalog.metadata.total_ru} RU")

pipeline = ValidationPipeline(Path(".claude/skills"))
report = pipeline.report(pipeline.run_all())
print(f"Errors: {report['errors']}, Warnings: {report['warnings']}")

CLI

Commands: stats, search, validate, quality, catalog.

agent-skills search <query>      # Search by name, description, tags
agent-skills validate --json out.json
agent-skills quality --json out.json
agent-skills catalog             # Rebuild catalog JSON
agent-skills stats               # Library statistics

GitHub Action

Validate, score, and catalog skills in your own repo with the Claude Skills Action:

name: Validate skills
on:
  pull_request:

jobs:
  validate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: ssrjkk/agent-skills@v3.5.0
        with:
          command: validate      # validate | quality | catalog | stats
          target: .claude/skills # path to skills directory
          output: skills_catalog.json

Quality Pipeline

Every skill is scored on 5 dimensions:

Domains (16)

ai · backend · blockchain · data · database · desktop · devops · embedded · engineering · finance · frontend · gamedev · healthcare · mobile · qa · security

Author

ssrjkk — creator & maintainer

For Contributors

See CONTRIBUTING.md. Quick checklist:

• SKILL.md has frontmatter with name, description, category, tags, models, version
• SKILL.ru.md is a real translation (not auto-generated)
• Code examples compile and run
• SKILL.md passes scripts/check_agent_interop.py (portable to all agents)
• make test passes
• make lint passes

Links

• Live Catalog Site
• Architecture Guide
• Release Notes
• Roadmap
• Issue Tracker

Why this library

• Curated, not generated — 100 hand-reviewed skills that pass a 5-dimension quality pipeline (completeness, depth, code quality, freshness, bilingual).
• Universal Agent Skills format — the same SKILL.md convention native to Claude Code and supported by OpenCode, Cursor, Windsurf, and others. No lock-in: install once, use anywhere.
• Bilingual (EN + RU) — every skill ships a real Russian translation, not a machine one.
• Works across every LLM/GLM ecosystem — skills are written to work with Claude (Sonnet/Opus), OpenAI (GPT), Google (Gemini), and Zhipu GLM models.
• Quality enforced in CI — validation, quality scoring, anti-pattern detection, cross-agent interop checks, 100% test coverage. What you see in main is what passes the bar.
• SDK + CLI + GitHub Action — validate, score, catalog and search any skills library programmatically.

Supported agents & models

The SKILL.md format is universal — every skill works with any agent that implements the Agent Skills spec, across model providers:

Legacy Version

The original v1.0 library (10,000+ auto-generated skills) is archived under the v1.0-legacy tag — v3.1 is a complete rewrite focused on quality:

git checkout v1.0-legacy

License

MIT.