Automatically learn from your workflows and turn them into intelligent, context-aware skills.

Auto-Skill v5.0 introduces Proactive Skill Discovery - a closed-loop learning system that observes your coding sessions, detects repeated patterns, and proactively suggests relevant community skills from 27,000+ external sources before you even ask. It combines local pattern detection with external skill discovery, cross-agent sharing, and anonymous telemetry.

npx skills add MaTriXy/auto-skill

npm install -g @matrixy/auto-skill

Once installed, the plugin automatically:

1. Records your tool usage patterns
2. Detects repeated workflows (3+ occurrences)
3. Offers to create skills from high-confidence patterns

auto-skill init                # Initialize config and directories
auto-skill discover            # Discover patterns for current project
auto-skill stats               # Show adoption statistics
auto-skill graduate            # Manage skill graduation
auto-skill agents list         # List known coding agents
auto-skill agents detect       # Detect installed agents
auto-skill lock status         # Show lock file status
auto-skill lock verify         # Verify skill integrity (SHA-256)
auto-skill lock list           # List locked skills
auto-skill telemetry report    # Show effectiveness report
auto-skill telemetry events    # Show raw events
auto-skill version             # Show version

All commands support --json output.

For skill search and installation, use Skills CLI:

npx skills find [query]        # Interactive skill search
npx skills add owner/repo      # Install skills from GitHub
npx skills list                # List installed skills

• Context-Aware Recommendations — Automatically suggests community skills based on detected frameworks, languages, and intent
• 27,000+ External Skills — Search and load from Skills.sh in real-time
• Hybrid Learning — Combines local pattern detection with external skill discovery
• Smart Graduation — Suggests upgrading local patterns to battle-tested community skills
• Zero-Config Search — Works out of the box, GitHub token optional for higher rate limits

• Pattern Detection — Automatically detects repeated tool sequences across sessions
• Session Analysis — Identifies intent (debug, implement, refactor) and workflow types (TDD, etc.)
• 18 Design Patterns — Architectural, coding, and workflow pattern recognition
• Mental Model Integration — Semantic codebase understanding via @mentalmodel/cli
• Multi-Agent Support — Cross-agent skill sharing (Claude Code, Codex, Cursor, Aider, etc.) via symlinks
• Provider System — Pluggable skill discovery (local, Skills.sh, RFC 8615 well-known endpoints)
• Lock File — SHA-256 integrity verification with atomic writes
• Spec Compliance — Generated skills validated against agentskills.io spec
• Path Security — Traversal prevention, null byte blocking, unicode normalization
• Confidence Evolution — Skills improve from 50% (external) → 75% (proven) → 85% (graduated)
• Anonymous Telemetry — Privacy-first usage tracking (details below)

import {
  createExternalSkillLoader,
  createProactiveDiscovery,
  createSkillRecommendationEngine,
} from '@matrixy/auto-skill';

const loader = createExternalSkillLoader({
  githubToken: process.env.GITHUB_TOKEN, // Optional
});

const discovery = createProactiveDiscovery(loader);
const engine = createSkillRecommendationEngine(loader, discovery);

await loader.start();

// Search community skills
const response = await loader.search('react testing', { limit: 5 });
console.log(response.skills);

// Get context-aware recommendations
const recommendations = await engine.recommendForPattern(detectedPattern);
console.log(recommendations);
// [
//   {
//     type: 'hybrid',
//     externalSkill: { title: 'React Test Patterns', ... },
//     localPattern: { ... },
//     reason: 'Your workflow matches "React Test Patterns". Use it instead?',
//     confidence: 0.9,
//     action: 'graduate'
//   }
// ]

await loader.stop();

See examples/proactive-discovery.ts for complete examples.

Full documentation: https://MaTriXy.github.io/auto-skill

• 🎉 Release Announcement — What's new in v5.0
• 📚 Proactive Discovery Guide — Complete usage guide
• 🎯 Example Code — 5 runnable examples

git clone https://github.com/MaTriXy/auto-skill.git
cd auto-skill
npm install
npm test

This tool collects anonymous, aggregate metrics (event types, counts, timing, OS) to improve the experience. No PII, search queries, file paths, or session data is collected.

Disable:

export AUTO_SKILL_NO_TELEMETRY=1  # or DO_NOT_TRACK=1

Automatically disabled in CI. Source: src/core/telemetry.ts

MIT License - see LICENSE

Version 5.0.0 | Repository | Issues | Changelog