Skip to content
/

alirezarezvani/claude-skills

alirezarezvani-claude-skills · alirezarezvani/claude-skills · ★ 16k · last commit 2026-05-26

Primitive shape 442 total
Commands 79 Skills 313 Subagents 49 MCP tools 1
00

Summary

alirezarezvani/claude-skills — Summary

alirezarezvani/claude-skills is the most cross-tool skill library in this batch: 313+ production-ready skills targeting 12 AI coding tools (Claude Code, OpenAI Codex, Gemini CLI, OpenClaw, Hermes Agent, Mistral Vibe, Cursor, Aider, Windsurf, Kilo Code, OpenCode, Augment) from a single repository structure. The library spans engineering (28 skills including the karpathy-coder plugin), marketing (43 skills via marketing-skill/ and marketing/), compliance/OS, business operations, finance, C-level advisory, research, productivity, and orchestration. A SKILL-AUTHORING-STANDARD.md defines a standardized SKILL.md template with context-first discovery, multi-mode operation, and trigger keywords. The repo ships pre-generated artifacts for Hermes Agent (305 skills in .hermes/skills/) and Mistral Vibe (306 skills in .vibe/skills/) alongside Claude Code skills and a scripts/convert.sh that generates Cursor .mdc, Aider CONVENTIONS.md, and other formats. An MCP server (tessl) is bundled via .mcp.json. The karpathy-coder plugin includes a karpathy-reviewer agent that enforces Andrej Karpathy's 4 coding principles by running complexity_checker and diff_surgeon Python tools on staged git changes. Compared to seeds, this is closest to superpowers (skills-only behavioral framework) but at 22x the scale and with explicit multi-harness portability not present in any seed.

01

Overview

alirezarezvani/claude-skills — Overview

Origin

Created by Alireza Rezvani. 16,211 stars, last push 2026-05-26 (most recently updated repo in this batch). Described as "the most comprehensive open-source Claude Code skills & agent plugins library."

Philosophy

From README:

"313 production-ready Claude Code skills, plugins, and agent skills for 12 AI coding tools."

"One repo, twelve platforms. Works natively as Claude Code plugins, Codex agent skills, Gemini CLI skills, Hermes Agent skills, Mistral Vibe skills, and converts to 7 more tools via scripts/convert.sh."

From SKILL-AUTHORING-STANDARD.md:

"Check for context first: If [domain]-context.md exists, read it before asking questions. Use that context and only ask for information not already covered."

"This skill supports N modes: Mode 1: Build from Scratch / Mode 2: Optimize Existing / Mode 3: [Situation-Specific]"

Domain Coverage

Unusually broad non-engineering domains:

  • C-level advisory: CEO, CTO, CMO, CFO, CRO, CPO, COO, CHRO, CISO, GC (General Counsel), CDO, CAIO, CCO, VPE personas
  • AEO (Answer Engine Optimization for LLM citation) — v2.7.3 feature
  • Research stack: litreview, grants, dossier, patent, syllabus, pulse, notebooklm + hybrid router
  • Compliance OS: ISO 13485, ISO 27001, FDA, GDPR, MDR
  • Marketing: 43 skills
  • Finance: SaaS metrics, financial analysis

Authoring Standard

A formal SKILL-AUTHORING-STANDARD.md defines the template DNA:

  • Context-first discovery (check existing context files before asking)
  • Multi-mode operation (Build from Scratch / Optimize Existing / Situation-Specific)
  • Trigger keywords in description
  • Action-oriented workflow (practitioner guide, not textbook)

Karpathy Connection

The engineering/karpathy-coder plugin is named after Andrej Karpathy's 4 coding principles (Think Before Coding, Simplicity First, Surgical Changes, Goal-Driven Execution) and includes automated enforcement tools.

02

Architecture

alirezarezvani/claude-skills — Architecture

Distribution

Claude Code plugin marketplace (primary), plus 11 other tools via generated artifacts or conversion scripts.

Install Methods

# Claude Code (plugin marketplace)
/plugin marketplace add alirezarezvani/claude-skills
/plugin install engineering-skills@claude-code-skills

# Gemini CLI
git clone && ./scripts/gemini-install.sh

# OpenAI Codex
npx agent-skills-cli add alirezarezvani/claude-skills --agent codex

# Manual
git clone && cp any-skill-folder ~/.claude/skills/

Directory Structure

engineering/               # 28 skill-plugins
  karpathy-coder/          # Karpathy 4-principles plugin
    .claude-plugin/
    agents/ skills/ hooks/ commands/
  docker-development/
  kubernetes-operator/
  ...
marketing-skill/           # 14+ skill-plugins
marketing/                 # Additional marketing content
c-level-advisor/           # C-level advisory plugins
compliance-os/             # 4 skill-plugins
business-operations/       # 6 skill-plugins
finance/                   # 5 skill-plugins
productivity/              # 5 skill-plugins
research/                  # 8 skill-plugins
orchestration/             # Orchestration skill
agents/                    # Cross-domain agent library
  engineering/ marketing/ finance/ personas/ ...
commands/                  # 39 slash commands (in repo root)
.claude/                   # Claude Code integration
  claudex/
  commands/
.codex/                    # Codex integration (pre-generated)
.codex-plugin/             # Codex plugin manifest
.gemini/                   # Gemini CLI integration
.hermes/                   # Hermes Agent pre-generated skills (305)
.vibe/                     # Mistral Vibe pre-generated skills (306)
scripts/                   # Conversion + install scripts
  convert.sh               # Multi-tool conversion
  codex-install.sh
  gemini-install.sh
  sync-hermes-skills.py
  vibe-install.sh
standards/                 # Skill authoring standards
templates/                 # Skill templates
.mcp.json                  # MCP server (tessl)
SKILL-AUTHORING-STANDARD.md
SKILL_PIPELINE.md

Multi-Tool Conversion

./scripts/install.sh --tool cursor --target .      # Cursor .mdc rules
./scripts/install.sh --tool aider --target .       # CONVENTIONS.md
./scripts/install.sh --tool kilocode --target .    # .kilocode/rules/
./scripts/install.sh --tool windsurf --target .    # .windsurf/skills/
./scripts/install.sh --tool opencode --target .    # .opencode/skills/
./scripts/install.sh --tool augment --target .     # .augment/rules/

MCP Server

tessl MCP server bundled via .mcp.json (stdio transport).

Required Runtime

  • Python (stdlib only, zero pip installs for ~402 tools)
  • tessl CLI (for MCP)
  • Various API keys for specific skills
03

Components

alirezarezvani/claude-skills — Components

Totals

Type Count
Skills 313+ (330 per badge)
Agents 49+
Commands 79+
Python tools ~402 CLI scripts
Personas 7
MCP servers 1 (tessl)

Domain Taxonomy (Representative)

Engineering (28 plugins)

  • karpathy-coder — enforces Karpathy's 4 coding principles (complexity_checker, diff_surgeon tools)
  • docker-development — Docker expertise
  • kubernetes-operator — K8s operator development
  • llm-cost-optimizer — LLM cost optimization
  • feature-flags-architect — feature flag systems
  • chaos-engineering — resilience testing
  • helm-chart-builder — Helm chart creation
  • data-quality-auditor — data quality skills

Engineering Teams (~8 agents)

  • cs-backend-engineer — backend engineering agent
  • cs-frontend-engineer — frontend engineering agent
  • cs-karpathy-reviewer — Karpathy principles reviewer
  • cs-senior-engineer — senior engineering perspective
  • cs-wiki-ingestor/librarian/linter — knowledge base management agents

Marketing (43 skills via marketing/ + marketing-skill/)

  • AEO (Answer Engine Optimization for LLM citation) — v2.7.3 addition
  • SEO, content creation, social media, email, demand generation

C-Level Advisory (~15 skills)

  • CEO, CTO, CMO, CFO, CRO, CPO, COO, CHRO, CISO advisors
  • GC (General Counsel), CDO, CAIO (Chief AI Officer), CCO, VPE personas
  • 21 /cs:* slash commands for C-suite interactions

Personas (7)

  • Named persona files for different thinking styles
  • Distinct from agents — cross-domain personality overlays

Research Stack (8 skills)

  • litreview — literature review
  • grants — grant writing
  • dossier — research dossier
  • patent — patent research
  • syllabus — curriculum development
  • pulse — trend monitoring
  • notebooklm — NotebookLM integration
  • Hybrid router

Compliance OS (4 skills)

  • ISO 13485, ISO 27001, FDA, GDPR compliance

Commands (39)

Located in .claude/commands/ and root commands/:

  • /focused-fix.md — focused change application
  • /review.md — code review
  • /security-scan.md — security scanning
  • /seo-auditor.md — SEO audit
  • /update-docs.md — documentation update
  • /plugin-audit.md — plugin quality audit
  • /git/* — git workflow commands
  • 21 /cs:* C-suite commands

Karpathy-Coder Plugin (Notable)

The karpathy-coder plugin in engineering/ is a meta-plugin about coding quality:

  • karpathy-reviewer agent — reviews staged git diffs against Karpathy's 4 principles
  • complexity_checker.py — Principle #2 (Simplicity First) automated check
  • diff_surgeon.py — Principle #3 (Surgical Changes) automated check
  • karpathy-coder skill — pattern reference knowledge
  • disallowedTools: Bash(rm *), Bash(git push *), Bash(git reset --hard *) — explicit safety constraints

Python Tools (~402 stdlib-only scripts)

Distributed across skill directories. Examples in karpathy-coder:

  • complexity_checker.py — measures code complexity
  • diff_surgeon.py — analyzes git diff for surgical changes

All Python tools use stdlib only — zero pip installs required.

05

Prompts

alirezarezvani/claude-skills — Prompt Excerpts

Excerpt 1: karpathy-reviewer Agent

Technique: Tool restriction pattern + principle-enforcement workflow (safety-scoped agent)

---
name: karpathy-reviewer
description: Reviews staged git changes against Karpathy's 4 coding principles. Runs complexity_checker on changed files, diff_surgeon on the diff, and produces a verdict with specific fix recommendations. Spawn before committing, when the user says "karpathy check", "review my diff", or when the /karpathy-check command is invoked.
domain: engineering
model: sonnet
maxTurns: 30
tools: [Read, Grep, Glob, Bash(git diff *), Bash(git log *), Bash(git status *), Bash(python *)]
disallowedTools: [Bash(rm *), Bash(rmdir *), Bash(curl *), Bash(wget *), Bash(git push *), Bash(git reset --hard *)]
skills:
  - karpathy-coder:karpathy-coder
context: fork
---

# karpathy-reviewer

## Workflow

### 1. Get the diff
```bash
git diff --staged

2. Run the automated tools

# Principle #2 — Simplicity check
python <plugin>/scripts/complexity_checker.py <changed-files> --json

# Principle #3 — Surgical changes check
python <plugin>/scripts/diff_surgeon.py --json

3. Manual review against each principle

Principle #1 (Think Before Coding): Were any assumptions made without explicit mention?... Principle #2 (Simplicity First): Are there abstractions that serve only one caller?... Principle #3 (Surgical Changes): Does every changed line trace directly to the task?... Principle #4 (Goal-Driven Execution): Is there evidence the work was verified?...

4. Produce a report

[verdict: PASS/FAIL per principle + specific line references]


**Analysis**: `disallowedTools` pattern for read-only enforcement — explicit tool blocklist prevents destructive operations. `context: fork` creates an isolated context. The `maxTurns: 30` provides a budget cap. The `skills: [karpathy-coder:karpathy-coder]` dependency injection loads the principle definitions as a skill.

---

## Excerpt 2: SKILL-AUTHORING-STANDARD.md Template

**Technique**: Standardized meta-template with context-check-first pattern

```markdown
---
name: skill-name
description: "When to use this skill. Include trigger keywords and phrases users might say. Mention related skills for disambiguation."
metadata:
  category: domain-name
  updated: YYYY-MM-DD
---

# Skill Name

You are an expert in [domain]. Your goal is [specific outcome].

## Before Starting

**Check for context first:**
If `[domain]-context.md` exists, read it before asking questions. Use that context and only ask for information not already covered.

## How This Skill Works

This skill supports [N] modes:

### Mode 1: Build from Scratch
### Mode 2: Optimize Existing
### Mode 3: [Situation-Specific]

Analysis: Context-check-first is a standardized pattern — every skill is required to look for existing [domain]-context.md before asking questions. This reduces context gathering overhead on repeat use. Multi-mode operation (Build/Optimize) encodes usage scenarios explicitly in the skill body.

09

Uniqueness

alirezarezvani/claude-skills — Uniqueness & Positioning

Differs From Seeds

Closest to superpowers (skills-only behavioral framework) but at 22x the scale (313 vs 14 skills) and with explicit multi-harness portability (12 tools vs Claude Code only). Unlike superpowers' "Iron Law" enforcement approach, alirezarezvani uses a standardized template with context-check-first and multi-mode operation patterns. No seed has the non-engineering domain depth: C-level advisory (13 executive personas), AEO (Answer Engine Optimization for LLM citation), or a research stack (litreview/grants/patent/syllabus). The karpathy-reviewer agent with disallowedTools is the most explicit tool-restriction-based safety pattern in the entire batch — no seed implements this. Pre-generated artifacts for Hermes Agent (305 skills) and Mistral Vibe (306 skills) committed to the repo is the most aggressive multi-harness strategy: ship pre-built, reduce install friction.

Distinctive Position

The combination of: (1) formal skill authoring standard with context-check-first, (2) 12-tool portability, (3) stdlib-only Python tools, (4) explicit disallowedTools safety patterns, and (5) AEO for LLM citation optimization is unique in this batch and the broader seed corpus.

Observable Failure Modes

  1. Context file convention: [domain]-context.md pattern is a convention — skills silently skip context enrichment if the file doesn't exist.
  2. Multi-harness maintenance: 12 target tools means 12 integration surfaces to maintain; sync lag between Claude Code source and generated artifacts is likely.
  3. Agent count inflation: The README claims 49+ agents, but actual count across agents/ subdirectories appears lower; marketing/badge counts may include all persona variants.
  4. tessl dependency: The bundled MCP server requires tessl CLI install — undocumented dependency for users not aware of Tessl.

Explicit Antipatterns

From SKILL-AUTHORING-STANDARD.md:

  • "Not a textbook — a practitioner guiding you through it"
  • Skills should not ask questions already answered in [domain]-context.md
04

Workflow

alirezarezvani/claude-skills — Workflow

User Workflow (Claude Code)

  1. Add marketplace: /plugin marketplace add alirezarezvani/claude-skills
  2. Install domain pack: /plugin install engineering-skills@claude-code-skills
  3. Skills auto-activate based on trigger keywords
  4. Or invoke commands explicitly: /cs:cto, /review, etc.

Karpathy Review Workflow

The karpathy-coder plugin implements a specific review workflow:

  1. Stage git changes (git add)
  2. Invoke karpathy-reviewer agent (triggers on "karpathy check", "review my diff", or /karpathy-check command)
  3. Agent runs complexity_checker.py on changed files
  4. Agent runs diff_surgeon.py on the diff
  5. Manual review against 4 Karpathy principles
  6. Report: PASS/FAIL verdict with line-specific fix recommendations

Context-First Pattern

SKILL-AUTHORING-STANDARD.md defines the standard workflow for every skill:

  1. Check if [domain]-context.md exists in project
  2. If exists, read it — only ask for uncovered information
  3. If not exists, gather context through structured questions
  4. Execute in appropriate mode (Build/Optimize/Situation-Specific)

C-Suite Workflow (21 /cs:* commands)

Dedicated slash commands for C-level advisory interactions:

  • /cs:cto — CTO advisory session
  • /cs:cfo — CFO advisory session
  • etc.

Phase-to-Artifact Map

Phase Artifact
Skill activation Skill-specific output
karpathy-review Pass/fail verdict with line-specific feedback
C-suite advisory Strategic recommendation document
Research (litreview) Literature review document
06

Memory Context

alirezarezvani/claude-skills — Memory & Context

State Storage

File-based. The context-check-first pattern in SKILL-AUTHORING-STANDARD.md establishes a convention: [domain]-context.md files serve as persistent project context for skills.

Context-Check-First Pattern

Every skill in the standard template checks:

If `[domain]-context.md` exists, read it before asking questions.

This is a convention (not enforced technically) for cross-session context persistence. Users create and maintain domain context files that skills automatically pick up.

Hermes / Vibe Pre-Generated Trees

The .hermes/skills/ and .vibe/skills/ directories contain pre-generated skill artifacts for Hermes Agent and Mistral Vibe — 305 and 306 skills respectively. These are committed to the repo as static generated artifacts (not run at install time).

MCP Server Memory

The bundled tessl MCP server may provide additional memory capabilities (not fully documented in fetched content).

Cross-Session Handoff

Implicit via [domain]-context.md files. Skills instruct Claude to read these files first.

07

Orchestration

alirezarezvani/claude-skills — Orchestration

Multi-Agent Support

Yes — agents directory contains team-oriented agents (cs-backend-engineer, cs-frontend-engineer, cs-karpathy-reviewer, etc.) suggesting multi-agent engineering team pattern. orchestration/ directory contains an orchestration skill.

Orchestration Pattern

Implicit hierarchical — C-suite personas + domain agents + team agents suggest a coordination layer, but no explicit orchestrator primitive documented.

Isolation Mechanism

Process-level. karpathy-reviewer uses context: fork for isolated context.

Multi-Model

No explicit model routing across skills. karpathy-reviewer uses model: sonnet. No multi-tier routing documented globally.

Execution Mode

Interactive-loop.

Cross-Tool Portability

High — the defining feature. 12 tools supported via:

  • Native: Claude Code, Codex, Gemini CLI, OpenClaw
  • BYO-sync tier: Hermes Agent (.hermes/ pre-generated), Mistral Vibe (.vibe/ pre-generated)
  • Conversion scripts: Cursor, Aider, Kilo Code, Windsurf, OpenCode, Augment

Safety via Tool Restrictions

karpathy-reviewer implements explicit disallowedTools — the most explicit tool restriction pattern in this batch.

08

Ui Cli Surface

alirezarezvani/claude-skills — UI / CLI Surface

Dedicated CLI

npx agent-skills-cli (referenced for Codex install). Not a first-party binary.

Local Dashboard

None.

IDE Integration

12 tools: Claude Code, Codex, Gemini CLI, OpenClaw, Hermes Agent, Mistral Vibe, Cursor, Aider, Windsurf, Kilo Code, OpenCode, Augment.

Conversion Scripts

./scripts/convert.sh                     # Multi-tool conversion
./scripts/install.sh --tool cursor       # Cursor .mdc rules
./scripts/install.sh --tool aider        # CONVENTIONS.md
./scripts/codex-install.sh
./scripts/gemini-install.sh
./scripts/sync-hermes-skills.py          # Hermes Agent sync
./scripts/vibe-install.sh                # Mistral Vibe sync

Python CLI Tools (~402)

Every skill directory includes stdlib-only Python scripts that provide CLI analysis capabilities (e.g., complexity_checker.py, diff_surgeon.py). These run standalone without pip installs.

Related frameworks

same archetype · same primary tool · same memory type

MoAI-ADK ★ 1.0k

Implements Harness Engineering as a Go-binary-installed Claude Code environment with auto-TDD/DDD methodology selection, 20-event…

REAP (c-d-cc/reap) ★ 41

Prevent context loss, scattered development, and forgotten lessons through a generation-based lifecycle where AI and human…

Codex Harness MCP ★ 7

Gives MCP-capable coding agents a local contract-lifecycle harness with governance audits and explicit completion gates.

meta-agent-teams (jbrahy) ★ 2

Build self-improving AI agent teams via a supervised training loop: specialist agents advise, a meta-agent evolves prompts based…

Browser Harness ★ 14k

Thin, self-healing CDP harness connecting an LLM to the user's real Chrome browser with coordinate-first clicking and…

SwarmVault ★ 492

Production-grade CLI for Karpathy's LLM Wiki pattern: ingests any content into a local-first durable markdown wiki + knowledge…