skill-check – skill Lập trình cho agent AI

SkillCheck

Overview

Validate SKILL.md files against the [agentskills specification](https://agentskills.io) and Anthropic best practices. Catches structural errors, semantic contradictions, naming anti-patterns, and quality gaps in a single read-only pass.

When to Use This Skill

• Use when user says "check skill", "skillcheck", or "validate SKILL.md"
• Use when reviewing a skill before publishing to a marketplace
• Use when debugging why a skill doesn't trigger correctly
• Use when onboarding a team to skill authoring standards
• Do NOT use for anti-slop detection, security scanning, or token analysis; use [SkillCheck Pro](https://getskillcheck.com) for those

How It Works

Step 1: Parse

Read the target SKILL.md file and extract YAML frontmatter.

Step 2: Validate

This file summarizes check categories but does not bundle a complete numbered ruleset or validator. Inspect the actual upstream rules/installed validator before claiming exact check IDs, counts or scores. With only this file, report a qualitative review by category and mark unavailable checks as not evaluated.

Apply available Free tier checks in order:

| Category | Checks | What it catches |

|----------|--------|----------------|

| Structure (1.x) | Name format, description WHAT+WHEN, allowed-tools, categories, XML injection | Malformed frontmatter, missing fields |

| Body (2.x) | Line count, hardcoded paths, stale dates, empty sections, deprecated syntax, MCP tool qualification | Content quality issues |

| Naming (3.x) | Vague terms, single-word names, gerund suggestions | Poor discoverability |

| Semantic (4.x) | Contradictions, ambiguous terms, missing output format, wisdom/platitudes, misplaced triggers | Logical inconsistencies |

| Quality (8.x) | Examples, error handling, triggers, output format, prerequisites, negative triggers | Strengths (positive patterns) |

Step 3: Score

Calculate overall score (0-100). Penalties: critical = -20, warning = -5, suggestion = -1.

Step 4: Report

Return structured results: score, grade (Excellent/Good/Needs Work/Poor), issue list with check IDs, line numbers, messages, and fix suggestions.

Examples

Example 1: Validating a skill

User: check my skill at ~/.claude/skills/weekly-report/SKILL.md

SkillCheck output:
## weekly-report Check Results [FREE]

Score: 85/100 (Good)

### Warnings (2)
  - 1.2-desc-when (line 3): Description missing WHEN clause
  - 4.5-desc-no-triggers (line 3): Description lacks triggering conditions

### Suggestions (1)
  - 3.4-gerund-naming (line 2): Skill name could use gerund form

### Passed Checks: 28

Example 2: Clean skill passes all checks

User: skillcheck ~/.claude/skills/processing-pdfs/SKILL.md

Score: 100/100 (Excellent)
All 31 checks passed. No issues found.

Limitations

• Read-only: does not modify any files
• Free tier covers structural, semantic, and naming checks only
• Anti-slop, security, WCAG, token, enterprise, and workflow checks require [SkillCheck Pro](https://getskillcheck.com)
• Semantic checks are heuristic; this catalog copy includes no evaluation evidence supporting a quantified false-positive rate.
• Does not validate referenced files or scripts; only checks SKILL.md content
• Single-file validation; does not cross-check against other skills in the same directory

Best Practices

• Run SkillCheck before submitting skills to any marketplace
• Fix all critical and warning issues; suggestions are optional
• Use the check ID (e.g., 1.2-desc-when) to find the exact rule in the skill body
• Re-run after fixes to confirm the score improved

Common Pitfalls

• Problem: Score seems low due to many suggestions

Solution: Suggestions cap at -15 points total. Focus on warnings and criticals first.

• Problem: False positive on ambiguous terms inside code blocks

Solution: SkillCheck skips code blocks and inline code. If you still see false positives, wrap the term in backticks.

• Problem: Wisdom/platitude check flags legitimate instructions

Solution: Rephrase generic advice ("Remember that testing is important") as concrete directives ("Run tests before committing").