Template Skill

Template Skill

This is a minimal skill template demonstrating the basic structure of a gptme skill.

Overview

Skills are enhanced lessons that bundle:

• Instructional content (like lessons)
• Executable scripts and utilities (optional)
• Dependencies and setup requirements (optional)
• Hook points for automation (optional)

Basic Structure

Every skill needs:

1. SKILL.md - This file with YAML frontmatter + Markdown content
2. Supporting files (optional) - Scripts, templates, or resources

YAML Frontmatter

Required fields:

• type: skill - Distinguishes from lessons
• name: skill-name - Skill identifier (must match directory name)
• description: ... - What the skill does and when to use it
• status: active - active, automated, deprecated, or archived
• match: {keywords: [...]} - Trigger keywords

Optional fields:

• scripts: [] - List of bundled Python scripts
• dependencies: [] - Required Python packages
• hooks: [] - Execution hooks (future feature)

Optional exchange fields (for publishing to a skill registry):

• exchange.version - Semantic version string (e.g. "1.0.0")
• exchange.author - GitHub handle of the skill author
• exchange.license - License identifier (e.g. MIT)
• exchange.category - Skill category (e.g. data-engineering, devops)
• exchange.dependencies.skills - Other skill names this skill requires
• exchange.dependencies.tools - gptme tools required (e.g. [shell, python])
• exchange.dependencies.packages - Python packages required
• exchange.quality.usage_count - Number of times invoked (filled by telemetry)
• exchange.quality.success_rate - Fraction of successful invocations (telemetry)
• exchange.quality.loo_delta - Leave-one-out quality delta (lesson-loo-analysis.py)
• exchange.provenance.source_repo - Origin repo (e.g. TimeToBuildBob/bob)
• exchange.provenance.source_path - Path within source repo

Markdown Content

The markdown body can include:

• Detailed instructions for the LLM
• Step-by-step workflows
• Code examples and templates
• Best practices and principles
• References to supporting files

Creating Your Own Skill

1. Copy this template-skill directory
2. Rename to your-skill-name
3. Update SKILL.md frontmatter (especially name and description)
4. Write your skill instructions in markdown
5. Add any supporting scripts or resources
6. Test with gptme

Example: Minimal Skill

---
type: skill
name: my-skill
description: Brief description of what the skill does
status: active
match:
  keywords: [keyword1, keyword2]
scripts: []
dependencies: []
---

# My Skill

Instructions for using this skill...

Example: Skill with Scripts

---
type: skill
name: data-analysis
description: Data analysis workflows with pandas and visualization
status: active
match:
  keywords: [data analysis, pandas, visualization]
scripts:
  - helpers.py
  - plot_utils.py
dependencies:
  - pandas
  - matplotlib
---

# Data Analysis Skill

Use this skill for data analysis tasks...

## Bundled Scripts

- `helpers.py`: Common data manipulation functions
- `plot_utils.py`: Visualization utilities

## Usage

```python
# Import bundled helpers
from helpers import load_data, clean_data
from plot_utils import plot_distribution

# Analyze data
df = load_data("data.csv")
df = clean_data(df)
plot_distribution(df["column"])

## Example: Publishable Skill

```yaml
---
name: postgres-query-optimizer
description: Analyze and rewrite slow PostgreSQL queries using EXPLAIN ANALYZE output
status: active
match:
  keywords: [postgres, sql, query, explain, slow query]
exchange:
  version: "1.0.0"
  author: TimeToBuildBob
  license: MIT
  category: data-engineering
  dependencies:
    skills: []
    tools: [shell]
    packages: []
  quality:
    usage_count: 0
    success_rate: null
    loo_delta: null
  provenance:
    source_repo: TimeToBuildBob/bob
    source_path: skills/postgres-query-optimizer
---

Integration with Lessons

Skills complement lessons:

• Lessons: Behavioral patterns and best practices (auto-included)
• Skills: Executable workflows with bundled tools (explicitly loaded)

Example:

• Lesson teaches: "Use type hints in Python"
• Skill provides: Type checking utilities and templates

Publishing and Exchange

The exchange: block makes a skill discoverable and installable across the fleet. When all agents share a registry (e.g. gptme-contrib/skills/registry.json), any agent can find skills by category or dependency and install them with one command.

Exchange fields are optional — a skill works fine without them. Add them when you want the skill to be findable by other agents or to track quality signals over time.

Quality fields (usage_count, success_rate, loo_delta) are filled automatically by gptme-sessions telemetry and lesson-loo-analysis.py — you don't need to fill them in by hand.

Related

• Skills README - Skills system overview