Agent Skills

A model-agnostic collection of skills for AI coding agents. Works with Claude Code, Cursor, Windsurf, and other AI-powered tools.

See also: agentskills.io - The community hub for AI agent skills

Author: ThepExcel • License: MIT

---

Installation Methods

Method 1: Plugin Marketplace (Recommended)

Easiest way - works with claude plugin install:

claude plugin install ThepExcel/thepexcel-claude-plugins

This installs from the thepexcel-claude-plugins repo (auto-synced from this repo).

Method 2: Manual Installation

These skills follow the Open Agent Skills specification and work with multiple AI coding agents.

Claude Code

# User-level (all projects)
git clone https://github.com/ThepExcel/agent-skills.git
mkdir -p ~/.claude/skills
cp -r agent-skills/deep-research ~/.claude/skills/
rm -rf agent-skills

# Project-level (current project only)
git clone https://github.com/ThepExcel/agent-skills.git
mkdir -p .claude/skills
cp -r agent-skills/deep-research .claude/skills/
rm -rf agent-skills

OpenAI Codex CLI

# User-level (all projects)
git clone https://github.com/ThepExcel/agent-skills.git
mkdir -p ~/.codex/skills
cp -r agent-skills/deep-research ~/.codex/skills/
rm -rf agent-skills

# Project-level (current project only)
git clone https://github.com/ThepExcel/agent-skills.git
mkdir -p .codex/skills
cp -r agent-skills/deep-research .codex/skills/
rm -rf agent-skills

Google Gemini CLI

Note: Agent Skills in Gemini CLI is experimental. Enable it first: /settings → search "Skills" → toggle on.

# User-level (all projects)
git clone https://github.com/ThepExcel/agent-skills.git
mkdir -p ~/.gemini/skills
cp -r agent-skills/deep-research ~/.gemini/skills/
rm -rf agent-skills

# Project-level (current project only)
git clone https://github.com/ThepExcel/agent-skills.git
mkdir -p .gemini/skills
cp -r agent-skills/deep-research .gemini/skills/
rm -rf agent-skills

Install Multiple Skills at Once

git clone https://github.com/ThepExcel/agent-skills.git

# Claude Code
mkdir -p ~/.claude/skills
cp -r agent-skills/deep-research agent-skills/triz agent-skills/problem-solving ~/.claude/skills/

# Codex CLI
mkdir -p ~/.codex/skills
cp -r agent-skills/deep-research agent-skills/triz agent-skills/problem-solving ~/.codex/skills/

# Gemini CLI
mkdir -p ~/.gemini/skills
cp -r agent-skills/deep-research agent-skills/triz agent-skills/problem-solving ~/.gemini/skills/

rm -rf agent-skills

Installation Scopes

CLI	User Scope	Project Scope
Claude	~/.claude/skills/	.claude/skills/
Codex	~/.codex/skills/	.codex/skills/
Gemini	~/.gemini/skills/	.gemini/skills/

---

Available Skills

Skill	Description
deep-research	Web research with Graph-of-Thoughts, hypothesis testing, source triangulation, Red Team critique
triz	TRIZ systematic innovation methodology for technical/physical contradictions and cross-industry solutions
explain-concepts	Educational explanations using Feynman, Socratic, and Cognitive Load methodologies
problem-solving	Systematic problem-solving with Good Teacher mode, inspired by Polya's "How to Solve It"
generate-creative-ideas	Creative ideation using SCAMPER, First Principles, Random Word, and AI-optimized techniques
design-business-model	Business model design with Business Model Canvas, Lean Canvas, and Value Proposition Canvas
manage-business-strategy	Strategic analysis, decision making, goal setting, and organizational change frameworks
power-query-coaching	Coaches data transformation in Power Query UI step-by-step
optimize-prompt	Prompt optimization for Claude, OpenAI, Gemini, n8n, and other AI platforms
extract-expertise	Extracts domain expertise from experts and transforms it into working Claude skills
create-visualization	Diagrams and animations for STEM education using Matplotlib and Manim
graphic-designer	Graphic design for thumbnails, social media, banners using CRAP and Gestalt principles
geometric-elements	Decorative geometric elements (corners, lines, frames) with gradient and transparency control
skill-creator-thepexcel	Guide for creating and enhancing Claude skills with audit rubric scoring

---

For AI Agents: Claude Code Complete Reference

This section is designed for AI agents (Claude, GPT, etc.) to understand Claude Code's architecture and use it effectively.

Table of Contents

1. Configuration Scope System
2. CLAUDE.md System
3. Skills vs Slash Commands
4. Creating Skills
5. Creating Slash Commands
6. MCP (Model Context Protocol)
7. Hooks System
8. Plugin Marketplace
9. File Organization Best Practices
10. New Commands & Features

---

1. Configuration Scope System

Claude Code uses a hierarchical configuration system with four scopes:

Scope	Location	Applies To	Version Control
Managed	System directories (see below)	All users on machine	Yes (deployed by IT)
User	~/.claude/	All projects globally	No (personal)
Project	.claude/ in repo root	Current project only	Yes (shared)
Local	.claude/settings.local.json	Current project, personal	No (gitignored)

Managed Settings Locations:

• macOS: /Library/Application Support/ClaudeCode/
• Linux: /etc/claude-code/
• Windows: C:\Program Files\ClaudeCode\

Precedence Order (highest to lowest)

1. Managed - Cannot be overridden (enterprise policies)
2. Command line arguments - Temporary session overrides
3. Local - Personal project overrides
4. Project - Team-shared settings
5. User - Personal defaults

Key Files by Scope

~/.claude/                          # USER SCOPE
├── settings.json                   # User preferences, permissions
├── CLAUDE.md                       # Global instructions for all projects
├── skills/                         # User-installed skills
├── commands/                       # User slash commands
├── agents/                         # User subagents
└── rules/                          # User-level rules (v2.0.64+)

.claude/                            # PROJECT SCOPE (version controlled)
├── settings.json                   # Project settings (includes hooks)
├── CLAUDE.md                       # Project-specific instructions
├── skills/                         # Project skills
├── commands/                       # Project slash commands
├── agents/                         # Project subagents
└── rules/                          # Modular rule files (v2.0.64+)

.claude/settings.local.json         # LOCAL SCOPE (gitignored)
CLAUDE.local.md                     # Personal project instructions (gitignored)

settings.json Structure

{
  "permissions": {
    "allow": ["Bash(git:*)", "Bash(npm *)"],
    "deny": ["Bash(rm -rf:*)"],
    "defaultMode": "default"
  },
  "hooks": {
    "PreToolUse": [...]
  },
  "env": {
    "MY_VAR": "value"
  },
  "model": "claude-sonnet-4-20250514",
  "language": "thai",
  "attribution": {
    "commit": "Generated with AI\n\nCo-Authored-By: Claude <noreply@anthropic.com>",
    "pr": ""
  }
}

New Settings (v2.1+)

Setting	Purpose	Example
language	Response language	"japanese", "thai"
respectGitignore	Control @ file picker	true (default)
showTurnDuration	Hide "Cooked for Xm Xs"	false
attribution	Customize commit/PR bylines	See above
fileSuggestion	Custom @ search command	{"type": "command", "command": "..."}
autoUpdatesChannel	Release channel	"stable" or "latest"

---

2. CLAUDE.md System

CLAUDE.md files provide persistent instructions that Claude reads automatically at the start of every conversation.

File Locations (all are read, in order)

Location	Purpose	When Read
Enterprise path (see Scope System)	Organization-wide instructions	Always
~/.claude/CLAUDE.md	Global user instructions	Always
./CLAUDE.md or ./.claude/CLAUDE.md	Project root instructions	When in project
./CLAUDE.local.md	Personal project instructions	When in project (gitignored)
./subdirectory/CLAUDE.md	Directory-specific	When working in that directory

Imports (v2.0.67+)

CLAUDE.md files can import other files using @path/to/file syntax:

See @README.md for project overview and @package.json for npm commands.

# Additional Instructions
- Git workflow: @docs/git-instructions.md
- Personal prefs: @~/.claude/my-project-prefs.md

Modular Rules with .claude/rules/ (v2.0.64+)

Organize instructions into multiple files:

.claude/rules/
├── code-style.md      # Code style guidelines
├── testing.md         # Testing conventions
├── security.md        # Security requirements
└── frontend/
    ├── react.md
    └── styles.md

Path-specific rules with frontmatter:

---
paths:
  - "src/api/**/*.ts"
  - "lib/**/*.ts"
---

# API Development Rules
- All endpoints must include input validation
- Use standard error response format

Best Practices for CLAUDE.md

1. Keep it concise - ~150-200 instructions max for reliable following
2. Use structured format - Headers, tables, code blocks
3. Include three elements:
   • WHAT: Project structure, tech stack, file locations
   • WHY: Purpose, goals, context
   • HOW: Commands to run, coding conventions, verification steps

Extended Memory Options

Beyond the official CLAUDE.md system, there are community tools for enhanced memory:

Tool	Description
claude-mem	Plugin that auto-captures session context and injects into future sessions
Memory Bank pattern	Structured memory files adapted from Cline methodology
MCP Memory servers	Knowledge graph via MCP for cross-project preferences
Simple docs/ folder	Store knowledge in docs/ and reference with @docs/file.md

Tip: Keep CLAUDE.md lean (loaded every session). Put detailed docs in separate files and use @imports or @docs/ references.

CLAUDE.md Template

# Project Name

## Context
Brief description of what this project does.

## Tech Stack
- Language: TypeScript
- Framework: Next.js
- Database: PostgreSQL

## Project Structure
\`\`\`
src/
├── components/    # React components
├── pages/         # Next.js pages
└── lib/           # Utility functions
\`\`\`

## Commands
- `npm run dev` - Start development server
- `npm test` - Run tests
- `npm run build` - Build for production

## Coding Conventions
- Use TypeScript strict mode
- Prefer functional components
- Write tests for new features

## Important Files
- `src/lib/api.ts` - API client
- `src/config.ts` - Configuration

---

3. Skills vs Slash Commands

Post v2.1.3: Just Use Skills

Key Change: As of v2.1.3, skills and slash commands are merged internally. They work identically — both appear in / menu, both can be auto-triggered by Claude.

Recommendation: Always Create Skills

Reason	Explanation
Same effort	Minimal skill = just skills/name/SKILL.md (one folder + one file)
More flexible	Can add scripts, references, assets later without restructuring
Cross-platform	Works on Claude.ai, API, AND Claude Code
Future-proof	Skills are the direction Anthropic is investing in

# Minimal skill structure
skills/my-task/
└── SKILL.md    # Same content as a command file would have

When Slash Commands Still Make Sense

Slash commands (commands/name.md) are only useful for:

• Legacy projects that already have them
• Quick throwaway scripts you'll delete soon

For new work: just create skills.

Comparison (for reference)

Aspect	Skills	Slash Commands
Location	skills/name/SKILL.md	commands/name.md
Supporting files	Yes	No
Cross-platform	Claude.ai, API, Claude Code	Claude Code only
Invocation	/name or auto	/name or auto
Recommendation	Use this	Legacy only

---

4. Creating Skills

Using the Official Skill-Creator

The easiest way to create skills is using the official skill-creator from Anthropic's plugin marketplace:

# Install from official Anthropic marketplace (auto-available)
/plugin install skill-creator@claude-plugins-official

# Then ask Claude to help create a skill
> Help me create a skill for [your use case]

Alternatively, you can install from the Anthropic Skills repo:

/plugin marketplace add anthropics/skills
/plugin install skill-creator@anthropic-skills

Skill Directory Structure

skills/
└── my-skill/
    ├── SKILL.md              # Required: main instructions
    ├── SOURCES.md            # Recommended: attribution
    ├── scripts/              # Optional: executable code
    │   └── helper.py
    ├── references/           # Optional: supporting docs
    │   ├── guide.md
    │   └── examples.md
    └── assets/               # Optional: templates, images
        └── template.txt

SKILL.md Format

---
name: skill-name
description: Describe when Claude should use this skill. Be specific about triggers.
allowed-tools:
  - Read
  - Grep
  - Glob
model: claude-sonnet-4-20250514
context: fork
agent: Explore
hooks:
  PreToolUse:
    - matcher: "Bash"
      hooks:
        - type: command
          command: "./validate.sh"
          once: true
---

# Skill Name

## When to Use
- User asks to "create something"
- User needs help with "specific task"

## Instructions

Step-by-step instructions for Claude to follow.

### Step 1: Understand the Request
Gather necessary information...

### Step 2: Execute the Task
Perform the main action...

### Step 3: Verify Results
Confirm the output is correct...

## Examples

### Example 1: Basic Usage
User: "Help me with X"
Action: Do Y, then Z

## Guidelines
- Always verify before executing
- Ask for clarification if ambiguous

SKILL.md Frontmatter Fields

Field	Required	Description
name	Yes	Lowercase, hyphens, max 64 chars
description	Yes	Trigger description, max 1024 chars. Claude uses this to decide when to apply
allowed-tools	No	Tools Claude can use without permission. Supports YAML lists
model	No	Override model (e.g., claude-sonnet-4-20250514)
context	No	Set to fork to run in isolated sub-agent context
agent	No	Agent type for fork: Explore, Plan, general-purpose, or custom
hooks	No	Scoped hooks: PreToolUse, PostToolUse, Stop
user-invocable	No	Show in / menu (default: true). Set false to hide
disable-model-invocation	No	Block Skill tool from calling this (default: false)

New Features (v2.1+)

• Hot-reload: Skills are reloaded automatically when modified (no restart needed)
• YAML-style lists: Use YAML lists for allowed-tools for cleaner syntax
• Progressive disclosure: Put detailed docs in separate files, Claude reads on-demand
• Bundled scripts: Scripts in skill directory can be executed without reading into context

---

5. Creating Slash Commands

Recommendation: Create Skills instead. Since v2.1.3, skills and commands work identically, but skills are more flexible and cross-platform. This section is kept for reference and legacy support.

Command File Location

.claude/commands/my-command.md      # Project scope
~/.claude/commands/my-command.md    # User scope

Command File Format

---
description: Brief description shown in /help
argument-hint: <required-arg> [optional-arg]
allowed-tools: Bash, Read, Write
---

## Context

Current information Claude should know:
- Git status: !`git status --short`
- Current branch: !`git branch --show-current`

## Your Task

Based on the context above, perform these steps:

1. First step
2. Second step
3. Final step

Do not do anything else.

Dynamic Context with !command

Use !backticks`` to inject command output (requires allowed-tools with Bash):

## Context
- Current directory: !`pwd`
- Files changed: !`git diff --name-only`
- Package version: !`node -p "require('./package.json').version"`

Positional Arguments

Access individual arguments with $1, $2, etc. (not just $ARGUMENTS):

---
argument-hint: [pr-number] [priority] [assignee]
description: Review pull request
---

Review PR #$1 with priority $2 and assign to $3.

Frontmatter Fields

Field	Required	Description
description	Yes	Shown in command help
argument-hint	No	Usage hint: <required> [optional]
allowed-tools	No	Restrict tools: Bash(git:*), Read
model	No	Override model for this command
context	No	Set to fork to run in sub-agent context
agent	No	Agent type for fork (requires context: fork)
hooks	No	Scoped hooks for this command's execution
disable-model-invocation	No	Block Skill tool from calling this

---

6. MCP (Model Context Protocol)

MCP allows Claude Code to connect to external tools and services.

Configuration File Locations

Scope	Location	Purpose
User	~/.claude.json	Personal MCP servers (in main config)
Project	.mcp.json	Shared MCP servers (version controlled)
Managed	/etc/claude-code/managed-mcp.json	Enterprise MCP servers

.mcp.json Format

{
  "mcpServers": {
    "stdio-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@package/mcp-server"],
      "env": {
        "API_KEY": "${API_KEY}",
        "HOST": "${HOST:-localhost}"
      }
    },
    "http-server": {
      "type": "http",
      "url": "https://api.example.com/mcp"
    }
  }
}

MCP Server Types

Type	Use Case	CLI Flag
stdio	Local process (default)	--command, --args
http	Remote HTTP API	--transport http
sse	Server-Sent Events	--transport sse

Environment Variable Expansion

• ${VAR} - Use environment variable
• ${VAR:-default} - Use default if not set

Adding MCP Servers via CLI

# Add stdio server (default)
claude mcp add github --command "npx" --args "-y @anthropic/mcp-github"

# Add HTTP server (v2.1+)
claude mcp add linear --transport http https://mcp.linear.app/mcp

# Add SSE server (v2.1+)
claude mcp add asana --transport sse https://mcp.asana.com/sse

# Add with JSON config
claude mcp add-json db '{"command":"npx","args":["-y","@anthropic/mcp-postgres"],"env":{"DATABASE_URL":"${DATABASE_URL}"}}'

# List servers
claude mcp list

# Enable/disable servers
/mcp enable server-name
/mcp disable server-name

# Remove server
claude mcp remove github

MCP Permissions (v2.0.70+)

Wildcard permissions for MCP tools:

{
  "permissions": {
    "allow": ["mcp__github__*"],
    "deny": ["mcp__filesystem__*"]
  }
}

MCP Settings

Setting	Purpose
enableAllProjectMcpServers	Auto-approve all project MCP servers
enabledMcpjsonServers	List of approved servers: ["github", "linear"]
disabledMcpjsonServers	List of blocked servers
allowedMcpServers	Managed allowlist (enterprise)
deniedMcpServers	Managed denylist (enterprise)

---

7. Hooks System

Hooks execute code in response to Claude Code events.

Hook Configuration Location

Important: Hooks are configured in settings.json, NOT in a separate hooks.json file.

Scope	Location
User	~/.claude/settings.json
Project	.claude/settings.json
Local	.claude/settings.local.json
Plugin	hooks/hooks.json (for plugins only)

Hook Configuration Format

In settings.json:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "python3 .claude/hooks/validate.py",
            "timeout": 30
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "hooks": [
          {
            "type": "command",
            "command": "$CLAUDE_PROJECT_DIR/.claude/hooks/format.sh"
          }
        ]
      }
    ],
    "SessionStart": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "echo 'Session started'"
          }
        ]
      }
    ]
  }
}

Hook Types (PascalCase!)

Hook	Trigger	Use Case
SessionStart	Session begins	Setup, logging, env init
SessionEnd	Session ends	Cleanup, summary
PreToolUse	Before tool execution	Validation, blocking, modification
PostToolUse	After tool execution	Logging, formatting, cleanup
PermissionRequest	Permission prompt shown	Auto-approve, policy enforcement
UserPromptSubmit	User sends message	Preprocessing, context injection
Stop	Claude stops responding	Post-processing
SubagentStop	Subagent completes	Aggregate results
PreCompact	Before context compaction	Save important context
Notification	Notification triggered	Custom notifications

Matcher Patterns

• Simple match: "Bash" - matches Bash tool only
• Regex: "Write|Edit" - matches Write or Edit
• Wildcard: "*" or "" - matches all tools
• MCP tools: "mcp__server__tool" format

Hook Options (v2.1+)

Option	Purpose
type	"command" (bash) or "prompt" (LLM-based)
command	Bash command to execute
prompt	LLM prompt (for type: "prompt")
timeout	Timeout in seconds (default: 600)
once	Run only once per session: true

Practical Example: Block .env File Access

Prevent accidental exposure of secrets by blocking direct reads of .env files:

// In settings.json
{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Read|Grep",
        "hooks": [
          {
            "type": "command",
            "command": "if echo \"$TOOL_INPUT\" | grep -qE '\\.env($|[^a-zA-Z])'; then echo '{\"decision\": \"block\", \"reason\": \"Direct .env access blocked. Use environment variables instead.\"}'; else echo '{\"decision\": \"approve\"}'; fi"
          }
        ]
      }
    ]
  }
}

Or use a script file .claude/hooks/block-env.sh:

#!/bin/bash
# Block direct access to .env files
if echo "$TOOL_INPUT" | grep -qE '\.env($|[^a-zA-Z])'; then
    echo '{"decision": "block", "reason": "Direct .env access blocked for security. Use environment variables instead."}'
else
    echo '{"decision": "approve"}'
fi

Then reference it:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Read|Grep",
        "hooks": [
          {
            "type": "command",
            "command": ".claude/hooks/block-env.sh"
          }
        ]
      }
    ]
  }
}

Hook Response (for PreToolUse)

Hook script can output JSON to control behavior:

{"decision": "block", "reason": "Operation blocked by policy"}

{"decision": "approve"}

{"decision": "ask", "updatedInput": {"command": "modified command"}}

Environment Variables

Variable	Description
$CLAUDE_PROJECT_DIR	Project root directory
${CLAUDE_PLUGIN_ROOT}	Plugin directory (for plugin hooks)

Hooks in Skills/Commands (v2.1+)

Define scoped hooks in frontmatter:

---
name: secure-deploy
hooks:
  PreToolUse:
    - matcher: "Bash"
      hooks:
        - type: command
          command: "./validate-deploy.sh"
          once: true
---

---

8. Plugin Marketplace

Installing from Marketplace

# Add marketplace (supports GitHub, GitLab, local paths, URLs)
/plugin marketplace add owner/repo-name
/plugin marketplace add https://gitlab.com/company/plugins.git
/plugin marketplace add ./local-marketplace

# Install plugin (user scope by default)
/plugin install plugin-name@marketplace-name

# Install with scope (-s or --scope)
/plugin install plugin-name@marketplace-name -s project
/plugin install plugin-name@marketplace-name -s user

# Enable/disable plugins
/plugin enable plugin-name@marketplace-name
/plugin disable plugin-name@marketplace-name

# Update marketplace
/plugin marketplace update marketplace-name

# Remove plugin
/plugin uninstall plugin-name@marketplace-name

Plugin Scopes

Scope	Who can use	Stored in
User	You, all projects	~/.claude/settings.json
Project	All collaborators	.claude/settings.json
Local	You, this project	.claude/settings.local.json
Managed	All users (enterprise)	Managed settings

Creating a Marketplace

Repository structure:

my-marketplace/
├── .claude-plugin/
│   └── marketplace.json
├── skills/
│   ├── skill-one/
│   │   └── SKILL.md
│   └── skill-two/
│       └── SKILL.md
├── plugins/                    # Mixed-component plugins
│   └── my-plugin/
│       ├── .claude-plugin/
│       │   └── plugin.json
│       ├── commands/           # NOT inside .claude-plugin/
│       ├── agents/
│       ├── skills/
│       ├── hooks/
│       │   └── hooks.json      # Plugin hooks go here
│       ├── .mcp.json           # Plugin MCP servers
│       └── .lsp.json           # Language server config (v2.0.74+)
└── README.md

Common Mistake: Don't put commands/, agents/, skills/, or hooks/ inside .claude-plugin/. Only plugin.json goes there.

LSP Support in Plugins (v2.0.74+)

Plugins can include Language Server Protocol configuration for code intelligence:

// .lsp.json
{
  "go": {
    "command": "gopls",
    "args": ["serve"],
    "extensionToLanguage": {
      ".go": "go"
    }
  },
  "python": {
    "command": "pyright-langserver",
    "args": ["--stdio"],
    "extensionToLanguage": {
      ".py": "python"
    }
  }
}

Users must have the language server binary installed.

marketplace.json Format

{
  "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
  "name": "marketplace-name",
  "version": "1.0.0",
  "description": "Description of this marketplace",
  "owner": {
    "name": "Owner Name",
    "email": "email@example.com"
  },
  "plugins": [
    {
      "name": "plugin-name",
      "description": "What this plugin does",
      "source": "./plugins/plugin-name",
      "category": "productivity",
      "tags": ["tag1", "tag2"],
      "skills": [
        "./skills/skill-one",
        "./skills/skill-two"
      ]
    }
  ]
}

plugin.json Format (for individual plugins)

{
  "name": "plugin-name",
  "description": "Plugin description",
  "version": "1.0.0",
  "author": {
    "name": "Author Name"
  }
}

---

9. File Organization Best Practices

Recommended Project Structure

project/
├── .claude/                    # Claude Code config (version controlled)
│   ├── settings.json           # Project settings + hooks
│   ├── settings.local.json     # Personal settings (gitignored)
│   ├── CLAUDE.md               # Alternative location for project instructions
│   ├── commands/
│   │   └── deploy.md
│   ├── skills/
│   │   └── project-skill/
│   │       └── SKILL.md
│   ├── agents/
│   │   └── reviewer.md
│   └── rules/                  # Modular rules (v2.0.64+)
│       ├── code-style.md
│       └── testing.md
├── .mcp.json                   # MCP servers (version controlled)
├── CLAUDE.md                   # Project instructions
├── CLAUDE.local.md             # Personal project instructions (gitignored)
└── src/                        # Project source code

.gitignore Recommendations

# Claude Code local settings
.claude/settings.local.json
.claude.local/

# Personal skills (if using symlinks)
.claude/skills/*
!.claude/skills/project-specific/

Naming Conventions

Component	Convention	Example
Skills	lowercase-with-hyphens	deep-research
Commands	lowercase-with-hyphens	deploy-prod
Agents	lowercase-with-hyphens	code-reviewer
Config files	lowercase.json	settings.json

---

10. New Commands & Features (v2.1+)

New Slash Commands

Command	Purpose
/plan	Enter plan mode directly from prompt
/teleport	Resume remote session from claude.ai
/remote-env	Configure remote session environment
/tasks	View and manage background tasks
/stats	Usage stats, streaks, model preferences
/rename <name>	Name sessions for easy resume
/config	Open settings interface (searchable)
/terminal-setup	Setup Shift+Enter for various terminals
/context	Visualize context window usage

Background Tasks (v2.0.60+)

• Press Ctrl+B to background running tasks
• Use /tasks to view background tasks
• Agents run asynchronously and notify when done

Named Sessions (v2.0.64+)

# Name current session
/rename my-feature

# Resume by name
claude --resume my-feature

# Or from REPL
/resume my-feature

Status Line (v2.0.65+)

Configure a custom status line in settings.json:

{
  "statusLine": {
    "type": "command",
    "command": "~/.claude/statusline.sh"
  }
}

Available fields: context_window.used_percentage, context_window.remaining_percentage, current_usage

Sandbox Mode

Isolate bash commands with filesystem and network restrictions:

{
  "sandbox": {
    "enabled": true,
    "autoAllowBashIfSandboxed": true,
    "excludedCommands": ["docker", "git"],
    "network": {
      "allowLocalBinding": true
    }
  }
}

---

Repository Structure

This repository (agent-skills) contains skills in a flat structure for easy cloning:

agent-skills/
├── deep-research/
│   ├── SKILL.md
│   ├── SOURCES.md
│   └── references/
├── triz/
├── ... (other skills)
└── README.md

The companion repository thepexcel-claude-plugins contains the same skills in nested plugin format for claude plugin install. It is auto-synced from this repo via GitHub Actions.

---

Contributing

1. Fork this repository
2. Create a new skill folder with SKILL.md
3. Add SOURCES.md for attribution
4. Submit a pull request

See the skill-creator-thepexcel skill for guidance on creating effective skills.

---

License

• ThepExcel original skills: MIT License
• skill-creator-thepexcel: Apache 2.0 (based on Anthropic's skill-creator)

---

Author

Created by Sira Ekabut (ThepExcel)

Developed through human-AI collaboration using Claude Code.