manage-skills

Managing Claude Skills

Before Creating a Skill

Where does this skill belong?

Firstloop shared skill (default assumption):

• Skills for core or monotemplate plugins
• Improvements to existing shared skills
• → Create a GitHub issue in firstloophq/claude-code-plugins (see meta-skill-management.md)

Local project skill:

• Project-specific workflows
• Custom skills for a single codebase
• User explicitly requests a "local skill"
• → Create directly in .claude/skills/<skill-name>/SKILL.md

Quick Reference

Scenario	Action
"Create a skill for monotemplate"	GitHub issue
"Improve the crud skill"	GitHub issue
"Add a docker skill to the plugin"	GitHub issue
"Create a local skill for this project"	Create in .claude/skills/
"Add a skill just for this repo"	Create in .claude/skills/

Creating Skills

Quick Start

Create a skill by making a SKILL.md file in .claude/skills/<skill-name>/:

---
name: my-skill-name
description: Brief description of what this skill does and when to use it.
---

# My Skill Name

## Instructions
[Your instructions here]

Required Structure

Every skill needs:

1. SKILL.md file with YAML frontmatter
2. name field: lowercase letters, numbers, hyphens only (max 64 chars)
3. description field: what the skill does AND when to use it (max 1024 chars)

Skill Directory Layout

.claude/skills/
└── my-skill/
    ├── SKILL.md          # Main instructions (required)
    ├── reference.md      # Additional details (optional)
    └── scripts/          # Utility scripts (optional)
        └── helper.py

Writing the Description

The description is critical for skill discovery. Include:

• What the skill does
• When Claude should use it
• Key trigger words

Good example:

description: Generate git commit messages by analyzing staged changes. Use when committing code, writing commit messages, or reviewing diffs.

Bad example:

description: Helps with git stuff

Skill Templates

Template 1: Instructions-Only Skill

For guidance-based skills without code:

---
name: code-review
description: Review code for bugs, style, and best practices. Use when reviewing pull requests, analyzing code quality, or checking for issues.
---

# Code Review

## Process

1. Read the code thoroughly
2. Check for bugs and edge cases
3. Verify style consistency
4. Suggest improvements

## Checklist

- [ ] No obvious bugs
- [ ] Error handling present
- [ ] Consistent naming
- [ ] No security issues

Template 2: Skill with Scripts

For skills that include executable utilities:

---
name: data-validation
description: Validate data files against schemas. Use when checking CSV, JSON, or config file formats.
---

# Data Validation

## Usage

Run validation:
```bash
python scripts/validate.py input.json schema.json

Scripts

• validate.py: Check data against schema
• format.py: Auto-fix formatting issues

See reference.md for schema format details.

### Template 3: Domain-Specific Skill

For skills with extensive reference material:

```markdown
---
name: api-integration
description: Integrate with the Acme API. Use when calling Acme endpoints, handling Acme webhooks, or processing Acme data.
---

# Acme API Integration

## Quick Reference

Base URL: `https://api.acme.com/v2`
Auth: Bearer token in header

## Common Operations

**Get user**: `GET /users/{id}`
**Create order**: `POST /orders`

For complete endpoint docs, see [endpoints.md](endpoints.md).
For authentication details, see [auth.md](auth.md).

Best Practices

Be Concise

Claude is smart. Only include information Claude doesn't already know.

Use Progressive Disclosure

• Keep SKILL.md under 500 lines
• Put detailed reference material in separate files
• Link with relative paths: [details](reference.md)

Set Appropriate Freedom

High freedom (guidelines):

Review the code for potential issues and suggest improvements.

Low freedom (specific steps):

Run exactly: `python scripts/migrate.py --verify --backup`
Do not modify the command.

Include Workflows for Complex Tasks

## Deployment Workflow

1. Run tests: `bun test`
2. Build: `bun run build`
3. If build fails, fix errors and repeat from step 1
4. Deploy: `bun run deploy`

Common Mistakes to Avoid

• Using vague descriptions
• Including time-sensitive information
• Using Windows-style paths (use forward slashes)
• Offering too many options without a default
• Deeply nesting file references (keep one level deep)

File Naming

Use descriptive, lowercase names with hyphens:

• api-reference.md (good)
• APIREF.md (bad)
• scripts/validate-schema.py (good)
• scripts/vs.py (bad)

Testing Your Skill

1. Create the skill files
2. Start a new conversation
3. Make a request that should trigger the skill
4. Verify Claude uses the skill correctly
5. Iterate based on behavior

---

Improving Existing Skills

When to Improve a Skill

• Skill produces inconsistent results
• Instructions are ambiguous or unclear
• Missing edge cases or workflows
• Description doesn't trigger skill when expected
• Skill has grown too large and needs refactoring

Improvement Process

1. Read the existing skill - Understand current structure and intent
2. Identify the problem - What's not working or missing?
3. Make targeted changes - Don't rewrite unless necessary
4. Test the changes - Verify improvements in a new conversation

Common Improvements

Sharpen the Description

If the skill isn't triggering when expected, improve the description:

# Before - too vague
description: Helps with database tasks

# After - specific triggers
description: Generate and run database migrations. Use when creating tables, modifying schemas, or updating database structure.

Add Missing Workflows

If Claude doesn't handle certain cases, add explicit workflows:

## Edge Cases

### Empty Input
If input is empty, prompt user for required data before proceeding.

### Existing File Conflict
If file already exists, ask user whether to overwrite, merge, or abort.

Reduce Ambiguity

Replace vague instructions with specific ones:

# Before - ambiguous
Review the code and suggest improvements.

# After - specific
Review the code for:
1. Security vulnerabilities (injection, XSS, auth issues)
2. Performance problems (N+1 queries, unnecessary loops)
3. Error handling gaps

For each issue found, provide:
- Location (file:line)
- Problem description
- Suggested fix

Split Large Skills

If a skill exceeds 500 lines, consider:

1. Moving reference material to separate files
2. Splitting into multiple focused skills
3. Using a reference folder structure:

my-skill/
├── SKILL.md           # Core instructions only
└── reference/
    ├── api.md         # API details
    ├── examples.md    # Extended examples
    └── errors.md      # Error handling guide

Improve Examples

Add before/after examples to clarify expected behavior:

## Examples

**Input**: User asks "format this JSON"
**Expected**: Format JSON with 2-space indentation, sorted keys

**Input**: User provides malformed JSON
**Expected**: Report specific syntax error location, suggest fix

Refactoring Checklist

• Description clearly states what and when
• Instructions are unambiguous
• Edge cases are documented
• Examples demonstrate expected behavior
• Reference material is in separate files
• File is under 500 lines

Additional Resources

• Frontmatter Rules - Detailed YAML validation rules
• Advanced Patterns - Feedback loops, workflows, scripts
• Meta Skill Management - How to contribute improvements to shared skills