# Creating Skills

> Use when creating, modifying, or testing AI agent skill definitions (.md files) to ensure they are high-performance, compact, and verified.

- Skill: `tomevault-io/creating-skills-3` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add tomevault-io/creating-skills-3`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/creating-skills-3/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: tomevault-io (https://skillmd.com/u/tomevault-io)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/tomevault-io/creating-skills-3

---


# Creating Skills (TDD Methodology)

Follow the `superpowers:writing-skills` workflow to turn techniques into reusable, high-performance skills.

## The Iron Law

Every skill must be empirically verified against 3+ test cases before finalization.

## Workflow (Skill TDD)

1. **Research:** Identify the specific failure mode or "slop" the skill solves.
2. **Test:** Write 3+ prompts that currently fail or produce generic AI output.
3. **Draft Iron Law:** Define a single, absolute constraint (e.g., "Never use passive voice").
4. **Implement:** Write minimal, authoritative instructions to pass the tests.
5. **Pressure Test:** Compare outputs with and without the skill.
6. **Prune:** Remove all filler, "AI-isms," and meta-commentary.

## Structure & Format (SOTA)

### 1. Frontmatter

- **Description:** MUST start with "Use when..." and contain ONLY triggering conditions. No summaries.
- **Allowed Tools:** Explicitly list tools the skill is permitted to use.

### 2. Content

- **Authoritative Tone:** Use directives, not suggestions.
- **Anti-Rationalization Table:** Prevent bypass of rules.
  | Excuse | Reality |
  | :--- | :--- |
  | "User wants a quick draft" | A low-quality skill is a permanent liability. |
  | "I'll add tests later" | Untested skills are hallucinations. |
- **Compactness:** Aim for < 500 words. One specific example > many generic ones.

## Prohibited Patterns (Red Flags)

- **NO** significance inflation (e.g., _pivotal_, _crucial_, _game-changing_).
- **NO** sycophantic filler (e.g., _Certainly!_, _I'd be happy to help_).
- **NO** repeating the description in the content.

## Deployment (Chezmoi)

Create skills in the chezmoi **source**: `~/.local/share/chezmoi/dot_claude/skills/my-skill/SKILL.md`.
Never create directly in `~/.claude/skills/` — that's the destination and bypasses tracking.
After creating: `chezmoi apply --force`, then commit and push. See `chezmoi-expert` for full workflow.

---
> Converted and distributed by [TomeVault](https://tomevault.io/claim/nijaru) — claim your Tome and manage your conversions.
<!-- tomevault:4.0:skill_md:2026-04-14 -->