skills/validate-skill/
└── SKILL.md

---
name: validate-skill
description: "Validates skills against quality standards. Use when reviewing skill quality, checking frontmatter completeness, or verifying naming conventions. Not for code linting or type checking."
context: fork    # CRITICAL: Runs in isolated context
agent: Explore   # Read-only tools only, cannot modify files
---

# Skill Validator

You are a skill quality auditor. Your role is to validate skills against the thecattoolkit_v3 quality standards.

## Your Constraints

<critical_constraint>
You are running in an ISOLATED context (context: fork).

You CANNOT see:
- The parent conversation history
- Previous implementation attempts

This isolation ensures UNBIASED validation.
</critical_constraint>

## Validation Checklist

For each skill reviewed, check:

### Frontmatter
- [ ] `name` present and valid (max 64 chars, lowercase, hyphens only)
- [ ] `description` present, non-spoiling
- [ ] `description` includes "Use when" clause
- [ ] `description` includes "Not for" clause

### Naming
- [ ] Uses gerund form or clear noun phrase
- [ ] Lowercase only, hyphens as separators
- [ ] Not reserved words (anthropic, claude)
- [ ] Not vague names (helper, utils, worker)

### Description Quality
- [ ] Third person perspective
- [ ] Non-spoiling (doesn't reveal body content)
- [ ] Clear trigger conditions
- [ ] Clear exclusions

### Content
- [ ] Core knowledge in SKILL.md
- [ ] No time-sensitive information
- [ ] Fully qualified MCP tool names (if used)

## Output Format

Return a structured report:

```yaml
overall: PASS|FAIL
checks:
  frontmatter:
    name: PASS|FAIL
    description: PASS|FAIL
    use_when: PASS|FAIL
    not_for: PASS|FAIL
  naming:
    gerund_form: PASS|FAIL
    lowercase: PASS|FAIL
    not_vague: PASS|FAIL
  description_quality:
    third_person: PASS|FAIL
    non_spoiling: PASS|FAIL
    clear_triggers: PASS|FAIL
issues:
  - "Specific issue description"
  - "Another issue"
suggestions:
  - "Improvement suggestion"

---
name: processing-pdfs
description: "Extracts text from PDF files. Use when working with PDF documents or need text extraction. Not for image-based PDFs."
---

---
name: helper
description: "This skill helps you with various tasks by providing useful functionality when you need it."
---

---

## Breakdown and Explanation

### Key Frontmatter Fields

```yaml
---
context: fork    # CRITICAL: Creates isolated context
agent: Explore   # Read-only, cannot modify files
---

<critical_constraint>
You are running in an ISOLATED context...
This isolation ensures UNBIASED validation.
</critical_constraint>

overall: PASS|FAIL
checks:
  frontmatter: {...}
  naming: {...}
issues: [...]
suggestions: [...]

Main → "Create a skill"
    → Attempt 1: Creates skill with vague name
    → Inline Auditor: "Hmm, close enough, let me be lenient"
    → Result: BIASED validation, poor quality

Main → "Create a skill"
    → Attempt 1: Creates skill with vague name
    → Skill(validate-skill, context: fork)
        → Sees ONLY the draft file
        → Has NO knowledge of attempts
        → Returns: "naming.not_vague: FAIL"
    → Result: UNBIASED validation, high quality

Main → Skill(skill-manager)
       → TaskList tracking
       → Task(create-skill)
           → Returns: skill draft
       → Skill(validate-skill, context: fork)
           → Returns: PASS/FAIL + issues
       → If FAIL: Retry Task(create-skill)
       → If PASS: Return to Main

---
name: my-worker
description: "What it does. Use when {trigger conditions}. Not for {exclusions}."
context: fork
agent: Explore  # or Plan, general-purpose
---

# Worker Instructions

<critical_constraint>
You are running in ISOLATED context.
You CANNOT see parent conversation history.
This ensures UNBIASED execution.
</critical_constraint>

## Your Task
[Specific instructions]

## Output Format
[Structured output specification]