Agent Skill Authoring and Testing
Drafts or fixes a SKILL.md with a trigger-ready description and lean rules, plus a with-vs-without test plan to prove it helps.
Before & after
Same prompt, same agent — the only difference is the skill.
To create a skill, you'll want to write a SKILL.md file that documents the skill's purpose, examples, and instructions. You should include sections on what it does, when to use it, and sample outputs. Make sure your instructions are clear and follow your team's conventions. You could then test it manually to ensure it works as intended. Once you're satisfied, you can deploy it into your Claude Code environment.
SKILL CREATION + VERIFICATION WORKFLOW STEP 1: Interview (gathering) -> Skill purpose: automate PostgreSQL migration generation and safety checks -> Trigger phrases: migrate, schema change, migration status, db-breaking -> Team conventions: migrations named YYYYMMDD_description, rollback scripts required -> Testing depth: unit test migrations + integration test on staging schema STEP 2: Write SKILL.md [File generated with sections: What You Get, When to Use, Example Output, Guardrails, Decision Tree] STEP 3: Subagent Verification -> Verification agent tests skill against 12 scenarios • Test 1: missing rollback script (should fail with guidance) • Test 2: breaking schema change without backfill (should block) • Test 3: valid migration with tests (should pass) • ...9 more edge cases -> Reports: PASS (11/12) or FAIL with details STEP 4: Output File: skills/postgres-migration-manager/SKILL.md Verification: PASS (11/12 tests) Ready to Deploy: yes Next: commit to skills/ and register in .claude/settings.json
About this skill
Your skill reads fine, but Claude never loads it. Or it loads on everything. Or it loads and the output is no better than plain Claude. This skill writes and debugs SKILL.md files with three goals: the description routes correctly, every rule in the body fixes a real mistake, and a head-to-head test plan shows the skill beats no skill.
What it does
- Writes the frontmatter description as a routing contract: what the skill does, how it does it, and real trigger phrases, including frustrated ones, all within the field limits.
- Runs a delta test on every sentence and cuts generic advice Claude already follows, so the rules that matter aren't buried.
- Fits each instruction to how fragile the task is: exact commands for fragile operations, heuristics with reasons for judgment work.
- Adds a size-matching section so the skill doesn't force a full template onto a one-line question.
- Diagnoses misfires: never triggers, always triggers, or overlaps with a neighboring skill. Fixes the description first.
- Builds a test plan: baseline failures, should-fire and should-not-fire prompts, and observable pass criteria for each.
What you get
A paste-ready SKILL.md, or ranked findings with replacement text for one you already have. You also get a short note on which baseline failure each key rule targets, and a with-skill vs without-skill test plan.
Best for
Developers and teams building Claude Code or Claude skills, marketplace sellers tightening a listing, and anyone whose existing skill is ignored, over-applied, or not clearly better than plain Claude.
SKILL.md preview
---
name: skill-authoring-and-testing
description: Writes, reviews, and repairs Claude Agent Skills (SKILL.md files). Produces a trigger-ready description, a body that contains only what Claude would get wrong without it, and a with-skill vs without-skill test plan covering should-fire, should-not-fire, and output quality. Use when the user says "write a skill for...", "turn this prompt into a SKILL.md", "review my skill", "my skill never triggers", "my skill fires on everything", "Claude ignores half my skill", "is this skill even helping?", or "split this skill up".
---
# Agent Skill Authoring and Testing
… (sign up to view the full skill)More ai & agents skills
View all AI & Agents skills →Autonomous /goal Prompt Builder
Turn rough ideas into one paste-ready /goal prompt with clear scope, constraints, acceptance checks, and autonomy rules.
Autonomous Creative Build Goal Prompts
Turn a rough game, 3D site, or video idea into one build prompt with scoped autonomy, acceptance tests, and delivery criteria.
Claude Power-User Coaching: One Tip Per Turn
Spots the Claude features and prompt habits you're missing in your actual work and coaches one concrete, copyable fix at a time.
Claude Skill Build and Launch
Turn an idea into a complete Claude skill with scoped triggers, acceptance tests, installation steps, and an agent launch plan.