Learner Skill

The Insight

Reusable skills are not code snippets to copy-paste, but principles and decision-making heuristics that teach the agent HOW TO THINK about a class of problems.

The difference:

• BAD (mimicking): "When you see ConnectionResetError, add this try/except block"
• GOOD (reusable skill): "In async network code, any I/O operation can fail independently due to client/server lifecycle mismatches. The principle: wrap each I/O operation separately, because failure between operations is the common case, not the exception."

A good skill changes how the agent APPROACHES problems, not just what code it produces.

Why This Matters

Before extracting a skill, ask yourself:

• "Could someone Google this in 5 minutes?" → If yes, STOP. Don't extract.
• "Is this specific to THIS codebase?" → If no, STOP. Don't extract.
• "Did this take real debugging effort to discover?" → If no, STOP. Don't extract.

If a potential skill fails any of these questions, it's not worth saving.

Recognition Pattern

Use /learner ONLY after:

• Solving a tricky bug that required deep investigation
• Discovering a non-obvious workaround specific to this codebase
• Finding a hidden gotcha that wastes time when forgotten
• Uncovering undocumented behavior that affects this project

The Approach

Extraction Process

Step 1: Gather Required Information

• Problem Statement: The SPECIFIC error, symptom, or confusion that occurred

  • Include actual error messages, file paths, line numbers
  • Example: "TypeError in src/hooks/session.ts:45 when sessionId is undefined after restart"

• Solution: The EXACT fix, not general advice

  • Include code snippets, file paths, configuration changes
  • Example: "Add null check before accessing session.user, regenerate session on 401"

• Triggers: Keywords that would appear when hitting this problem again

  • Use error message fragments, file names, symptom descriptions
  • Example: ["sessionId undefined", "session.ts TypeError", "401 session"]

• Scope: Almost always Project-level unless it's a truly universal insight

Step 2: Quality Validation

The system REJECTS skills that are:

• Too generic (no file paths, line numbers, or specific error messages)
• Easily Googleable (standard patterns, library usage)
• Vague solutions (no code snippets or precise instructions)
• Poor triggers (generic words that match everything)

Step 3: Save Location

• User-level: ~/.factory/skills/omc-learned/ - Rare. Only for truly portable insights.
• Project-level: .omd/skills/ - Default. Version-controlled with repo.

What Makes a USEFUL Skill

CRITICAL: Not every solution is worth saving. A good skill is:

1. Non-Googleable: Something you couldn't easily find via search

   • BAD: "How to read files in TypeScript" ❌
   • GOOD: "This codebase uses custom path resolution in ESM that requires fileURLToPath + specific relative paths" ✓

2. Context-Specific: References actual files, error messages, or patterns from THIS codebase

   • BAD: "Use try/catch for error handling" ❌
   • GOOD: "The aiohttp proxy in server.py:42 crashes on ClientDisconnectedError - wrap StreamResponse in try/except" ✓

3. Actionable with Precision: Tells you exactly WHAT to do and WHERE

   • BAD: "Handle edge cases" ❌
   • GOOD: "When seeing 'Cannot find module' in dist/, check tsconfig.json moduleResolution matches package.json type field" ✓

4. Hard-Won: Took significant debugging effort to discover

   • BAD: Generic programming patterns ❌
   • GOOD: "Race condition in worker.ts - the Promise.all at line 89 needs await before the map callback returns" ✓

Anti-Patterns (DO NOT EXTRACT)

• Generic programming patterns (use documentation instead)
• Refactoring techniques (these are universal)
• Library usage examples (use library docs)
• Type definitions or boilerplate
• Anything a junior dev could Google in 5 minutes

Skill Format

Skills are saved as markdown with this structure:

YAML Frontmatter

Standard metadata fields:

• id, name, description, source, triggers, quality

Body Structure (Required)

# [Skill Name]

## The Insight
What is the underlying PRINCIPLE you discovered? Not the code, but the mental model.
Example: "Async I/O operations are independently failable. Client lifecycle != server lifecycle."

## Why This Matters
What goes wrong if you don't know this? What symptom led you here?
Example: "Proxy server crashes on client disconnect, taking down other requests."

## Recognition Pattern
How do you know when this skill applies? What are the signs?
Example: "Building any long-lived connection handler (proxy, websocket, SSE)"

## The Approach
The decision-making heuristic, not just code. How should the agent THINK about this?
Example: "For each I/O operation, ask: what if this fails right now? Handle it locally."

## Example (Optional)
If code helps, show it - but as illustration of the principle, not copy-paste material.

Key: A skill is REUSABLE if the agent can apply it to NEW situations, not just identical ones.

Related Commands

• /note - Save quick notes that survive compaction (less formal than skills)
• /ralph - Start a development loop with learning capture