Skills System

Commander's skills system provides domain-specific expertise that agents can load on demand.

Architecture

skills/
├── skillManager.ts       ← Skill lifecycle management
├── skillCurator.ts       ← Skill curation and organization
├── skillInjector.ts      ← Inject skills into agent prompts
├── skillStore.ts         ← Persistent skill storage
├── skillQualityScorer.ts ← Quality scoring for skills
├── skillSecurityScanner.ts ← Security scanning for skill content
├── skillViewTool.ts      ← Tool for viewing/loading skills
├── metaLearnerBridge.ts  ← Bridge to MetaLearner for skill optimization
├── types.ts              ← Shared types
└── index.ts              ← Public exports

What is a Skill?

A skill is a packaged set of instructions, examples, and constraints that teaches an agent how to perform a specific type of task. Skills can be:

• Built-in — Shipped with Commander
• User-defined — Created and shared by users
• Community — Downloaded from skill registries
• Learned — Auto-generated by the MetaLearner

Managing Skills

# List available skills
npx tsx packages/core/src/cliEntry.ts skill list

# View a skill's content
npx tsx packages/core/src/cliEntry.ts skill view <skill-name>

# Create a new skill
npx tsx packages/core/src/cliEntry.ts skill create <skill-name>

# Pin a skill (always loaded)
npx tsx packages/core/src/cliEntry.ts skill pin <skill-name>

Skill Quality

Each skill is automatically scored on:

Metric	Description
Effectiveness	How often using the skill improves outcomes
Precision	How accurately the skill guides the agent
Completeness	Whether the skill covers all necessary aspects
Safety	No harmful or insecure instructions
Freshness	How recently the skill was updated

Security Scanning

Skills are automatically scanned for:

• Malicious instructions
• Data exfiltration attempts
• Prompt injection vulnerabilities
• Unsafe tool usage patterns

Creating a Skill

// skills/my-domain-skill.json
{
  "name": "domain-expert",
  "version": "1.0.0",
  "description": "Expert knowledge for domain X",
  "instructions": [
    "When working with domain X, always follow these patterns...",
    "Preferred tools for domain X are: read, search, analyze"
  ],
  "constraints": [
    "Never use write tool on production files",
    "Always validate schema before making changes"
  ],
  "examples": [
    {
      "task": "analyze domain X component",
      "approach": "1. Read the component files\n2. Identify patterns\n3. Generate report"
    }
  ]
}