validate-claude-tools

Validate Claude AI tools against Anthropic's official documentation and best practices.

Target: $ARGUMENTS (if empty, validate all of .claude/)

Setup

1. If $ARGUMENTS is provided, validate only that path
2. If $ARGUMENTS is empty, validate the entire .claude/ directory
3. Create output file: CLAUDE_MEMO_AI_TOOLS_REVIEW.md in the project root

Frontmatter Schemas

IMPORTANT: Each tool type has different valid frontmatter fields. Use these authoritative schemas.

Subagents (.claude/agents/*.md)

---
name: lowercase-with-hyphens        # Required (max 64 chars)
description: What this agent does   # Required (max 1024 chars)
tools: Read, Write, Bash            # Optional - tools the agent can use
model: sonnet | opus | haiku | inherit  # Optional - defaults to inherit
permissionMode: default | acceptEdits | bypassPermissions | plan | ignore  # Optional
skills: skill1, skill2              # Optional - skills to auto-load
---

Skills (.claude/skills/*/SKILL.md)

---
name: skill-name                    # Required (max 64 chars)
description: What and when          # Required (max 1024 chars)
allowed-tools: ["Read", "Write"]    # Optional - constrain tool access
---

Commands (.claude/commands/*.md)

---
description: What this command does # Required
argument-hint: <arg1> [arg2]        # Optional
allowed-tools: Read, Write          # Optional
model: sonnet | opus | haiku        # Optional - override model for this command
---

Key distinction: Agents use tools: field. Skills and commands use allowed-tools: field. Do not confuse these.

---

Validation Framework

For each tool type, check the applicable rules below.

Skills (.claude/skills/*/SKILL.md)

Structure validation:

• SKILL.md exists in skill directory
• YAML frontmatter has opening and closing --- delimiters
• Required field: name (lowercase, numbers, hyphens only, max 64 chars)
• Required field: description (non-empty, max 1024 chars, no XML tags)
• No reserved words in name: "anthropic", "claude"