Agent skill

writing-skills

Methodology for creating effective skills using TDD principles. Use when creating new skills, editing existing skills, or authoring plugin components. Triggers on: 'create skill', 'write skill', 'new skill', 'skill authoring'.

Stars 122
Forks 4

Install this agent skill to your Project

npx add-skill https://github.com/TechDufus/oh-my-claude/tree/main/plugins/oh-my-claude/skills/writing-skills

SKILL.md

Writing Skills Skill

TDD applied to process documentation.

Core Principle

Writing skills IS TDD applied to process documentation. If you can't test it breaks without the skill, the skill isn't needed.

When to Apply

  • Creating new skills
  • Editing existing skills
  • Reviewing skill quality
  • Authoring plugin components

TDD Mapping

TDD Concept Skill Equivalent
Test case Pressure scenario (situation where agent fails without skill)
Production code SKILL.md content
Failing test Agent demonstrably fails the scenario
Passing test Agent handles scenario correctly with skill loaded
Refactor Trim to minimum effective content

RED-GREEN-REFACTOR for Skills

RED: Identify Failure Scenarios

Identify 3+ pressure scenarios where the agent fails without this skill. Document the failure mode. This is your test suite. If you can't find 3 scenarios, the skill probably isn't needed.

GREEN: Write Minimum Effective Content

Write SKILL.md that addresses every identified failure. Minimum content to pass -- no extras, no nice-to-haves, no "comprehensive guides." Every section must map to at least one pressure scenario.

REFACTOR: Trim Aggressively

Remove anything that doesn't directly address a pressure scenario. Target ~120 lines. If removing a line doesn't weaken any pressure scenario, remove it.

SKILL.md Structure Template

---
name: {short-name}
description: "{Role sentence}. Use when {triggers}. Triggers on: {keywords}."
---
# {Name} Skill
{Tagline}
## When to Apply
## Core Pattern / Framework
## Anti-Rationalization Table (if methodology skill)
## Common Mistakes
## The Bottom Line

Claude Search Optimization (CSO)

The description field determines when Claude loads your skill. Get it wrong and the skill never fires.

  • Description = "Use when..." triggers ONLY
  • NEVER summarize the workflow in the description
  • Include trigger keywords that match how users/agents naturally phrase the need

Good: "Use when receiving review comments from code-reviewer agent, PR reviews, or external feedback. Triggers on: 'review feedback', 'address review', 'fix review comments'."

Bad: "A comprehensive guide to handling code review feedback through a structured process"

The bad example describes what the skill contains. The good example describes when to load it.

Skill Types

Type Purpose Example Line Target
Technique How to do X debugger ~120 lines
Pattern When to use X receiving-code-review ~120 lines
Methodology Discipline for X tdd, verification ~130 lines
Meta How to create X writing-skills ~140 lines

Anti-Rationalization for Skipping Tests

Excuse Counter
"The skill is obviously clear" If it's obvious, you can write 3 failure scenarios in 2 minutes
"It's just a reference doc" Reference docs that don't address failures are shelf-ware
"Testing documentation is overkill" Testing docs IS identifying what problems they solve
"I'll test it in practice" Without defined scenarios, you won't notice when it fails

When NOT to Create a Skill

Not everything deserves a skill:

  • Information easily found in docs (link to docs instead)
  • One-time procedures (just do them)
  • Tool-specific configuration (use CLAUDE.md)
  • Content under 30 lines (too simple -- put it in CLAUDE.md)

Token Efficiency

Every line must earn its place:

  • Technique/pattern skills: ~120 lines
  • Methodology skills: ~130 lines
  • Meta skills: ~140 lines

The test: "If removing this line doesn't weaken any pressure scenario, remove it."

Avoid:

  • Repetitive examples that demonstrate the same point
  • "Nice to know" sections that don't prevent failures
  • Long prose when a table communicates faster
  • Headers with only 1-2 lines of content (merge them)

Frontmatter Rules

Skills use ONLY name and description in frontmatter:

yaml
---
name: my-skill
description: "What it does. Use when X. Triggers on: 'keyword1', 'keyword2'."
---

No model, tools, or permissionMode. These are not agent definitions.

Common Mistakes

Mistake Why It's Wrong Do This Instead
Writing without pressure scenarios No way to verify the skill works RED phase first -- always
Summarizing workflow in description Claude can't match it to triggers Use "Use when..." pattern
Including everything you know Bloats context, dilutes signal Only what prevents identified failures
Skipping refactor phase Skills grow stale and verbose Trim after every edit
Copying another skill's structure Different types need different shapes Match structure to skill type

The Bottom Line

A skill without tested pressure scenarios is documentation. Documentation without a problem to solve is noise.

Expand your agent's capabilities with these related and highly-rated skills.

TechDufus/oh-my-claude

tdd

Systematic TDD methodology for writing tests first. Use when implementing features or fixes with OMC_TDD_MODE enabled, when unsure how to structure tests, or when test-first discipline needs reinforcement. Triggers on: 'tdd', 'test first', 'test driven', 'red green refactor'.

122 4
Explore
TechDufus/oh-my-claude

pr-creation

MUST be used when creating pull requests. Handles context gathering, title generation (conventional commit format), body formatting, and PR creation via GitHub CLI. Creates PRs as drafts by default. Triggers on: 'create PR', 'open PR', 'ready for review', 'push for PR', 'send for review'.

122 4
Explore
TechDufus/oh-my-claude

ralph-loop-init

Transform approved plans into ralph loop infrastructure. Triggers on: '/ralph-loop-init', '/ralph-init', 'setup ralph loop', 'generate ralph loop'. Creates .ralph/ directory with prd.json, loop.py, CLAUDE.md, and supporting files.

122 4
Explore
TechDufus/oh-my-claude

ralph-plan

Structured PRD generation with interview, research, and approval workflow. Triggers on: '/ralph-plan <topic>', 'create prd', 'generate prd', 'plan this'. Creates comprehensive Product Requirements Document via interview and research.

122 4
Explore
TechDufus/oh-my-claude

verification

Evidence-based verification methodology before claiming work is complete. Use when about to claim work is done, before committing, before creating PRs, or when verification-reminder hook fires. Triggers on: 'verify', 'verification', 'is it done', 'complete', 'ready to merge'.

122 4
Explore
TechDufus/oh-my-claude

git-commit-validator

MUST be used for ANY git workflow that involves committing code. This includes explicit commit requests AND implicit ones like 'ship it', 'wrap it up', or finishing implementation work. Handles staging, message generation, validation, and commit execution with conventional commit format.

122 4
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results