write-skill

name	write-skill
description	Use when creating a new skill, editing an existing skill, or verifying a skill works before deployment.
user-invocable	true
argument-hint	[skill-name]
allowed-tools	Read, Write, Edit, Bash, Glob, Grep

/write-skill - Create or Improve Skills with TDD

Writing skills IS Test-Driven Development applied to process documentation.

Target: $ARGUMENTS

The Iron Law

NO SKILL WITHOUT A FAILING TEST FIRST

If you didn't verify an agent fails without the skill, you don't know if the skill teaches the right thing.

Violating the letter of this rule is violating the spirit of this rule.

Skill Structure

skills/
  my-skill/
    SKILL.md              # Main skill file (required)
    supporting-file.*     # Only if needed (heavy reference, tools)

SKILL.md Template

---
name: my-skill
description: "Use when [specific triggering conditions and symptoms]"
user-invocable: true
argument-hint: "[arguments]"
allowed-tools: Read, Write, Edit, Bash, Glob, Grep
---

# /my-skill - Title

[Core principle in 1-2 sentences.]

**Target:** $ARGUMENTS

## When to Use

- [Symptom or situation 1]
- [Symptom or situation 2]
- NOT for: [when not to use]

## Workflow

### Step 1: [Action]
[Instructions]

### Step 2: [Action]
[Instructions]

## Verification Protocol

**Before claiming skill execution is complete:**
[What commands to run as proof]

## Anti-Rationalization

| Excuse | Reality |
|--------|---------|
| "[common excuse]" | [why it's wrong] |

## Rules

1. [Rule 1]
2. [Rule 2]

## Output

[Expected output format]

CSO (Claude Search Optimization)

Critical for discovery. Future Claude reads the description to decide if it should load the skill.

Description Rules

1. Start with "Use when..." - Focus on triggering conditions
2. NEVER summarize the workflow - Claude will follow the summary and skip the content
3. Include symptoms - Error messages, situations, contexts
4. Third person - Descriptions are injected into system prompts
5. Under 500 characters - Be concise

# BAD: Summarizes workflow - Claude may follow this instead of reading skill
description: "Debug issues using 4-phase methodology: gather, analyze, hypothesize, fix"

# GOOD: Just triggering conditions
description: "Use when encountering any bug, test failure, or unexpected behavior - before proposing fixes"

Name Rules

• Use only lowercase letters, numbers, and hyphens
• Use verb-first, active voice: write-skill not skill-writing
• Be descriptive: condition-based-waiting not async-helpers

TDD Cycle for Skills

RED: Write Failing Test (Baseline)

Before writing the skill, test what happens WITHOUT it:

1. Create a scenario that the skill should handle
2. Ask an agent to handle it WITHOUT the skill loaded
3. Document EXACTLY what the agent does wrong:
   - What choices did it make?
   - What rationalizations did it use?
   - Where did it deviate from best practice?
4. This is your "failing test" - the baseline behavior

GREEN: Write Minimal Skill

Write a skill that addresses the specific failures from RED:

1. Create skills/[name]/SKILL.md
2. Address ONLY the failures observed in baseline
3. Don't add hypothetical cases
4. Test again WITH the skill loaded
5. Verify the agent now behaves correctly

REFACTOR: Close Loopholes

1. Look for NEW rationalizations the agent uses
2. Add explicit counters for each rationalization
3. Build the Anti-Rationalization table
4. Add Red Flags list
5. Re-test until the skill is bulletproof

Validation Checklist

Before deploying the skill:

### Structure
- [ ] YAML frontmatter with name and description
- [ ] Description starts with "Use when..."
- [ ] Description does NOT summarize workflow
- [ ] Name uses only lowercase, numbers, hyphens

### Content
- [ ] Clear overview with core principle
- [ ] Workflow with numbered steps
- [ ] Verification Protocol section
- [ ] Anti-Rationalization table
- [ ] Rules section
- [ ] Output format section

### Quality
- [ ] One excellent example (not multi-language)
- [ ] No narrative storytelling
- [ ] Concise (< 500 words for frequently-loaded skills)
- [ ] Keywords for search throughout

### Testing
- [ ] Baseline tested WITHOUT skill (RED)
- [ ] Agent complies WITH skill (GREEN)
- [ ] Rationalizations addressed (REFACTOR)

Anti-Rationalization

Excuse	Reality
"Skill is obviously clear"	Clear to you is not clear to other agents. Test it.
"It's just a reference"	References can have gaps. Test retrieval.
"Testing is overkill"	Untested skills always have issues. 15 min testing saves hours.
"I'll test if problems emerge"	Test BEFORE deploying, not after.
"Too tedious to test"	Testing is less tedious than debugging a bad skill in production.
"No time to test"	Deploying untested skill wastes more time fixing it later.

Output

──── /write-skill ────
Skill: [name]
Status: [created | updated]

Structure: ✓ Valid
CSO: ✓ Description starts with "Use when..."
Testing: [RED ✓ | GREEN ✓ | REFACTOR ✓]

File: skills/[name]/SKILL.md

Rules

1. No skill without failing test first - TDD applies to documentation too
2. CSO-optimize descriptions - "Use when..." not "What it does"
3. Never summarize workflow in description - Claude will skip the content
4. Address every rationalization - Build the table from real test failures
5. One excellent example - Beats five mediocre ones
6. Verify before deploying - Run the validation checklist