Auto-Skill

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

Auto-Skill observes your coding sessions across 10 supported agents, detects repeated patterns, and generates reusable SKILL.md files. Skills are automatically shared across all your installed agents. It combines local pattern detection with external skill discovery, cross-agent sharing, and anonymous telemetry.

Quick Start

npx skills add MaTriXy/auto-skill

Install CLI

# With pip
pip install aiskill

# With uv
uv pip install aiskill

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

CLI

auto-skill init                # Initialize config and directories
auto-skill discover            # Discover skills for current project
auto-skill search "query"      # Search external skills
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.

Key Features

• 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
• External Discovery — 27,000+ community skills from Skills.sh
• Mental Model Integration — Semantic codebase understanding via @mentalmodel/cli
• Multi-Agent Support — 10 coding agents (Claude Code, Codex, Cursor, Aider, etc.) with cross-agent skill sharing 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)

Documentation

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

Development

git clone https://github.com/MaTriXy/auto-skill.git
cd auto-skill
uv sync --all-extras
uv run pytest tests/ -v

Telemetry

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: core/telemetry.py

License

MIT License - see LICENSE

Version 3.0.2 | Repository | Issues | Changelog