Skill Creator

Skill Creator

Goal

Transform any newly learned capability into a permanent, reusable skill that persists across sessions.

When to Use

• After successfully completing a new type of task
• When user asks to "remember how to do this"
• When you've built a useful automation or integration
• After learning a new API or tool

Process

Step 1: Identify the Capability

Analyze what was just accomplished:

• What problem did this solve?
• What inputs were required?
• What outputs were produced?
• What steps were involved?

Step 2: Generalize the Pattern

Abstract from the specific instance:

• Remove hardcoded values
• Identify variable inputs
• Document decision points
• Note error handling patterns

Step 3: Create Skill Structure

skills/<skill-name>/
├── SKILL.md          # Main skill definition
└── reference.md      # Examples and detailed docs (optional)

Step 4: Write SKILL.md

Use this template:

---
name: skill-name-here
description: One-line description of what this skill does.
---

# Skill Name

## Goal
Clear statement of what this skill accomplishes.

## Process
1. First step of the process
2. Second step
3. Continue until complete

## Inputs
- Input one: description
- Input two: description

## Outputs
- Output one: description
- Where outputs are saved

## Key Principles
- Important rules
- Things to always/never do

## References
- Related skills or documentation

Step 5: Save and Register

• Save to skills/<skill-name>/SKILL.md
• For project-specific: .agent/skills/
• For global: ~/.agent/skills/

Step 6: Announce Acquisition

Tell the user:

"I've created a new skill: [skill-name]

This skill allows me to [capability description].

I can now do this anytime - it's permanently saved.

To use it: [trigger phrase or command]"

Skill Quality Checklist

Before saving, verify:

• Goal is clear and specific
• Process steps are actionable
• Inputs are well-defined
• Outputs are specified
• No hardcoded values that should be parameters
• Error cases are handled
• Can be understood by another agent

Example: Creating an API Integration Skill

After successfully integrating with a new API:

---
name: weather-api
description: Fetch weather data from OpenWeatherMap API.
---

# Weather API Integration

## Goal
Retrieve current weather and forecasts for any location.

## Process
1. Receive location (city name or coordinates)
2. Authenticate with API key from credentials
3. Call appropriate endpoint (current/forecast)
4. Parse response into structured data
5. Return formatted weather information

## Inputs
- Location: city name, zip code, or lat/lon
- Type: "current" or "forecast"
- Units: "metric" or "imperial" (default: metric)

## Outputs
- Temperature, conditions, humidity, wind
- For forecast: 5-day hourly breakdown
- Formatted for display or further processing

## Key Principles
- Cache responses for 10 minutes to avoid rate limits
- Handle API errors gracefully with retry
- Always include units in output

Self-Bootstrapping Pattern

This skill enables the autonomous agent loop:

┌─────────────────────────────────────────┐
│                                         │
│   User Request                          │
│        ↓                                │
│   Agent Attempts                        │
│        ↓                                │
│   Success? ─── No ──→ Research & Build  │
│        │                     ↓          │
│       Yes               Test & Iterate  │
│        │                     ↓          │
│        └────────────────────┘           │
│                 ↓                       │
│        [SKILL-CREATOR]                  │
│                 ↓                       │
│        Permanent Skill                  │
│                 ↓                       │
│        Announce Capability              │
│                                         │
└─────────────────────────────────────────┘

References

• See AUTONOMOUS_BOOTUP_SPEC.md for full architecture
• See memory-manager skill for persistence patterns