Claude's Code & Agent Plugins: 362 AI Agent Skills Library
/claude-skills
Verdict: A Vast Library for AI Agent Enhancement. Claude Code Skills & Agent Plugins compiles 362 domain-specific instruction packages, 644 pure-stdlib Python scripts, and 30+ agent personas under one MIT roof. It genuinely runs on 13 different AI coding tools without format conversion for a subset of them, which is technically useful. The tradeoff is navigability — with hundreds of skills across 18 domains, finding the right starting point is genuinely hard. Reach for this if you want domain expertise (regulatory compliance, clinical research, C-level advisory) that no single agent ships with out of the box. Skip it if you just need basic code generation; the setup surface area is larger than the job.
What Are AI Agent Skills and Plugins?
What Are AI Agent Skills and Plugins? Claude Code Skills & Agent Plugins is a modular instruction library that gives AI coding agents domain expertise they don't carry out of the box. Each skill is a SKILL.md file paired with optional Python CLI scripts and reference templates; skills are organized into bundles like engineering, marketing, regulatory compliance, or C-level advisory. The repo also defines agent personas (pre-configured identities with curated skill loadouts) and orchestration patterns for combining them. Everything runs on 13 AI coding tools after format conversion via `scripts/convert.sh`.
Core Features and Skill Categories
- ✓362 skills across 18 domains: engineering (core + ADVANCED tier), product, marketing, productivity, academic research, research operations, project management, regulatory/quality, compliance OS, C-level advisory, business operations, commercial, and finance
- ✓644 stdlib-only Python CLI scripts — zero pip installs required, works anywhere Python runs
- ✓30+ pre-built agent personas (Startup CTO, Growth Marketer, Solo Founder) with curated skill loadouts and distinct communication styles
- ✓Security auditor skill (skill-security-auditor) scans any skill for command injection, code execution, data exfiltration, and supply chain risks before installation
- ✓AEO (Answer Engine Optimization) skill tracks citation patterns across 5 LLMs and audits E-E-A-T signals
- ✓Multi-tool conversion via `scripts/convert.sh` — generates native format packages for Cursor, Aider, Windsurf, Kilo Code, OpenCode, Augment, and Antigravity from a single source tree
- ✓741 reference templates, checklists, and domain-specific knowledge files bundled alongside skills
- ✓Skill chaining and orchestration patterns — Solo Sprint, Domain Deep-Dive, Multi-Agent Handoff, and Skill Chain — defined in a lightweight protocol requiring no extra framework
Advanced Capabilities: The Ultimate Tier
- ✓Genuine multi-tool support: 13 AI coding agents with format conversion handled by a shell script, not manual work
- ✓The practical application tier of 25 advanced skills covers real engineering needs — RAG architect, database designer, MCP server builder, incident commander — with enough depth to avoid generic advice
- ✓Skill security auditor ships with the repo and scans skills for malicious code, code execution risks, and supply chain attacks before you install anything
- ✓741 bundled reference templates and checklists mean skills come with working artifacts, not just prompts
- ✓All 644 Python scripts use stdlib only — no dependency hell, no pip install step
- ✓The C-level advisory bundle (68 skills) with founder-mode agents and a full C-suite persona set covers a domain that no other open-source agent skill repo addresses
Limitations & risks
- △Discovery is a real problem: 362 skills in 18 domains with no built-in recommendation engine or interactive picker — browsing the directory tree is the only on-ramp
- △Quality is uneven across 362 packages; some are production-grade with scripts and references, others are thin prompts with minimal tooling, and the README doesn't grade them
- △Hermes Agent and Mistral Vibe require a one-time manual sync step per machine — not a blocker, but not friction-free either
- △The `convert.sh --tool all` step takes around 15 seconds per the README and regenerates files locally; CI environments need this accounted for in the pipeline
- △No version pins or changelog per skill — updating the library means reinstalling everything with no diff of what changed
- △The MIT license is permissive, but skills that wrap proprietary standards (FDA, ISO 27001, MDR) don't carry legal clearance — the repo provides workflow guidance, not regulatory guarantees
Quick Installation for Your AI Agent
For Claude Code, add the marketplace and install by domain bundle: `/plugin marketplace add alirezarezvani/claude-skills` then `/plugin install engineering-skills@claude-code-skills`. For Codex, run `npx agent-skills-cli add alirezarezvani/claude-skills --agent codex`. For multi-tool use, clone the repo and run `./scripts/convert.sh --tool all` to generate all 13 native formats locally, then `./scripts/install.sh --tool cursor --target /path/to/project` to inject them into a project. Hermes Agent and Mistral Vibe each ship a pre-generated skill tree and a one-time sync script (`python scripts/sync-hermes-skills.py` or `./scripts/vibe-install.sh`).
Combining Skills, Agents, and Personas
The repo distinguishes three building blocks: skills (how to execute a task), agents (what task to run), and personas (who is thinking). Skills are installed individually or as bundles; agents are pre-configured sessions with a goal; personas are identity presets with personality-driven communication styles. The recommended workflow is orchestration: start with a persona, stack relevant skills, then hand off to a second persona for review. For example, run `startup-cto + aws-solution-architect + senior-frontend` for two weeks of build work, then switch to `growth-marketer + launch-strategy + copywriting` for the launch phase. No framework is required — the protocol is documented in orchestration/ORCHESTRATION.md.
Exploring Related AI Agent Projects
Common Questions About Agent Skills
Run `/plugin marketplace add alirezarezvani/claude-skills` inside Claude Code, then install a skill bundle with `/plugin install engineering-skills@claude-code-skills` (or any named bundle from the README). Alternatively, clone the repo and copy skill folders manually to ~/.claude/skills/.
Yes — the library targets 13 tools: Claude Code, OpenAI Codex, Gemini CLI, OpenClaw, Hermes Agent, Mistral Vibe, Cursor, Aider, Windsurf, Kilo Code, OpenCode, Augment, and Antigravity. Run `./scripts/convert.sh --tool all` to generate native formats, then use `./scripts/install.sh --tool <name>` to inject them into a project.
MIT license. You can use, modify, and redistribute all 362 skills, 644 Python scripts, and 741 reference templates in commercial and personal projects, provided you include the original copyright notice.
Skills span 18 domains: engineering architecture, RAG pipeline design, security auditing, CI/CD generation, MDR/FDA compliance documentation, clinical research coordination, C-suite advisory, SEO and AEO optimization, SaaS financial modeling, and more. Each skill is a structured SKILL.md with optional Python CLI scripts and reference templates.
Skills answer 'how to execute a task' in a single domain. Agents are pre-configured sessions scoped to a specific job like 'run a security audit.' Personas define 'who is thinking' — they carry personality, communication style, and cross-domain context. Personas load multiple skills; skills can run without a persona. All three compose via the documented orchestration patterns.
Yes — the skill-security-auditor skill (v2.0.0+) scans any SKILL.md and its associated scripts for command injection, code execution, data exfiltration, prompt injection, and dependency supply chain risks. It returns PASS / WARN / FAIL with remediation guidance and requires no external dependencies.
The problem it solves
AI coding agents ship with general reasoning but lack domain depth. Asking a bare Claude Code or Codex session to run a clinical research protocol, draft MDR Annex II compliance docs, or architect a multi-agent RAG pipeline means writing custom prompts from scratch every time. There was no reusable library of structured agent instruction packages with supporting scripts — until this repo filled that gap.
Best use cases
- •Architecture review: load the senior-architect skill to identify top scalability risks in a microservices codebase
- •Compliance documentation: run the mdr-745-specialist skill to audit technical docs against MDR Annex II requirements
- •Content pipeline: chain the seo-audit skill with a copywriting skill to produce and optimize a blog post in one pass
- •Multi-agent product launch: cycle through startup-cto for architecture, growth-marketer for positioning, and solo-founder for iteration across a 6-week sprint
- •Clinical research coordination: invoke the clinical-research and research-finance skills together to scope a study design and its R&D budget simultaneously
- •Enterprise DevOps: use ci-cd-pipeline-builder to auto-generate GitHub Actions configs from a detected tech stack
Who should try it — and who should skip
Claude Code Skills & Agent Plugins is worth installing if you want domain-depth expertise (regulatory compliance, clinical research, RAG architecture, financial modeling) inside an AI coding agent and you're comfortable browsing a large library to find what you need. Solo founders and small teams who want Claude Code to behave like a full consultancy will get the most out of the C-level advisory personas and business skill bundles. If you just want AI-assisted code completion without setup overhead, the discovery and installation steps are overkill — use a bare agent first and revisit this when you need structured domain workflows.
Related repositories
Is claude-skills worth your time?
ChatGPT, Claude and Perplexity can all read this page. Ask one of them what it makes of claude-skills.