Quick Start
Once you've installed Auto-Skill and registered the hooks, here's what happens.
1. Work Normally
Use your coding agent as you always do. The observer hook silently records tool events (which tools were called, inputs, outcomes) into a local SQLite database.
Every tool call generates an event like:
{
toolName: "Read",
toolInput: { file_path: "src/auth.ts" },
success: true,
sessionId: "abc123",
projectPath: "/my-project"
}
2. Patterns Emerge
After a few sessions, the detector looks for repeated tool sequences. For example, if you consistently:
- Read a file
- Edit it
- Bash to run tests
- Bash to check output
...and this happens 3+ times across sessions, it becomes a pattern candidate.
3. Review Suggestions
Discover detected patterns:
auto-skill discover
Output:
Detected Patterns:
1. read-edit-test-verify (confidence: 0.82, occurrences: 5)
2. debug-read-grep-fix (confidence: 0.74, occurrences: 3)
4. Generate Skills
Patterns above the confidence threshold (default: 0.7) can be turned into SKILL.md files. The generated skill includes:
- YAML frontmatter with metadata and confidence scores
- Step-by-step workflow description
- Session context (intent, workflow type)
- Design patterns detected in your code
5. Discover Community Skills
Auto-Skill can proactively suggest community skills from Skills.sh when it detects patterns. For manual searching, use Skills CLI:
npx skills find testing workflow # Interactive search
npx skills add owner/repo@skill # Install specific skill
External skills start at 50% confidence and graduate to local skills after proving themselves (5+ uses, 80% success rate).
6. Web Dashboard (Optional)
Launch the web UI to visualize patterns and manage skills:
auto-skill start:web
# or: npm run start:web
# Open http://localhost:8000
What's Next
- Configuration — Tune detection thresholds and enable features
- How It Works — Understand the pipeline architecture
- Pattern Detection — Deep dive into how patterns are found