Skip to content
Back to skills

Verify Claude Setup

ASecurity

Verify .claude directory configuration is complete and correct. Use when checking if agents, hooks, rules, and memory are properly set up, or after making changes to .claude configuration.

  • 57 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added May 31, 2026
toolsbashtestingdatabasesecurity

Security analysis

A100/100

Scanned May 31, 2026

npx -y skills add diegosouzapw/awesome-omni-skill --skill verify-claude-setup --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Verify Claude Setup?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Verify Claude Setup
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/diegosouzapw-verify-claude-setup/badge)](https://www.skillsdirectory.com/skills/diegosouzapw-verify-claude-setup)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: verify-claude-setup
description: Verify .claude directory configuration is complete and correct. Use when checking if agents, hooks, rules, and memory are properly set up, or after making changes to .claude configuration.
allowed-tools: [Read, Bash, Grep, Glob]
---

# Verify .claude Setup

Quick verification that .claude directory is properly configured.

## When to Use

- After updating agents, rules, hooks, or memory
- Before starting complex work (ensure system is ready)
- User asks "is everything set up correctly?"
- Troubleshooting configuration issues

## Verification Checklist

### 1. Agents (should have 6)
```bash
ls .claude/agents/ | wc -l
```

Verify all agents have:
- Proper YAML frontmatter
- Research-First Protocol section
- Clear responsibilities

### 2. Hooks (should have 11 scripts + README)
```bash
ls .claude/hooks/scripts/*.ts | wc -l
```

Check:
- All scripts executable (`chmod +x`)
- All have Bun shebang (`#!/usr/bin/env bun`)
- settings.json has all hooks configured
- Pre-tool and Post-tool hooks exist

### 3. Rules (should have research-first-protocol.md)
```bash
ls .claude/rules/*.md
```

Verify:
- research-first-protocol.md exists
- Has `alwaysApply: true`

### 4. Memory (should have 8 files)
```bash
ls .claude/memory/ | wc -l
```

Check CLAUDE.md imports all:
- research-first-enforcement.md
- coding-standards.md
- testing-standards.md
- architecture-patterns.md
- common-workflows.md
- build-commands.md
- asset-forge-guide.md
- security-protocols.md

### 5. Settings
```bash
grep -c "hooks" .claude/settings.json
```

Verify:
- SessionStart, UserPromptSubmit hooks
- PreToolUse (Bash, Write, Edit) hooks
- PostToolUse (Write, Edit, Read, Grep, Glob) hooks
- PreCompact, Stop, SubagentStop, SessionEnd hooks

### 6. Commands (should have ~20)
```bash
find .claude/commands -type f -name "*.md" | wc -l
```

## Quick Verification

Run all checks:
```bash
echo "Agents: $(ls .claude/agents/ | wc -l)"
echo "Hooks: $(ls .claude/hooks/scripts/*.ts | wc -l)"
echo "Rules: $(ls .claude/rules/*.md | wc -l)"
echo "Memory: $(ls .claude/memory/ | wc -l)"
echo "Commands: $(find .claude/commands -type f -name "*.md" | wc -l)"
echo "Skills: $(ls .claude/skills/ | wc -l)"
```

## Expected Counts

- Agents: 6
- Hook Scripts: 11
- Rules: 2 (README + research-first-protocol)
- Memory: 8
- Commands: ~20
- Skills: ~40

## Report Format

Provide summary:
```
✅ Agents: 6/6 with Research-First Protocol
✅ Hooks: 11/11 executable, all configured
✅ Rules: research-first-protocol.md active
✅ Memory: 8/8 files, all imported in CLAUDE.md
✅ Commands: 20 organized slash commands
✅ Skills: 40 skills ready

Status: .claude configuration VERIFIED
```

Or if issues found:
```
⚠️ Agents: Missing Research-First Protocol in database-specialist.md
⚠️ Hooks: pre-tool-write.ts not executable
❌ Memory: research-first-enforcement.md not imported in CLAUDE.md

Status: Configuration INCOMPLETE - needs fixes
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.