Discovery & Graduation

Beyond local pattern detection, Auto-Skill connects to external skill sources and manages trust through an adoption system.

Unified Discovery

The UnifiedSuggester merges skills from multiple sources and ranks them:

Local Patterns  →  highest priority (actual usage data)
Proven External →  high priority (used and successful)
Mental Hints    →  medium priority (semantic match)
External Search →  lower priority (community, untested)

Searching

# Discover skills from all sources
auto-skill discover

# View adoption stats
auto-skill stats

# JSON output
auto-skill discover --json

To search the Skills.sh community directly, use Skills CLI:

npx skills find authentication    # Interactive search
npx skills add owner/repo@skill   # Install specific skill

Skills.sh Integration

Skills.sh hosts 27,000+ community-contributed skills. When hybrid.enable_external is true, Auto-Skill searches this registry for skills matching your detected patterns.

External skills are installed to ~/.claude/skills/external/ and start with a 50% confidence score.

Adoption Tracking

Every external skill is tracked in a local SQLite database:

Field	Description
usage_count	Times the skill was triggered
success_count	Times it led to a successful outcome
initial_confidence	Starting confidence (50% for external)
current_confidence	Evolves with usage
graduated_to_local	Whether it's been promoted

Confidence increases with successful usage and decreases with failures.

Graduation

When an external skill proves itself, it can graduate to a local skill with full confidence:

Criteria:

• 5+ uses
• 80% success rate
• Confidence reaches 85%

Process:

1. GraduationManager.detectCandidates() finds eligible skills
2. User reviews and approves the promotion
3. A local SKILL.md is generated in ~/.claude/skills/auto/
4. The graduation is logged in graduation_log.json
5. The skill tracker marks it as graduated

# View graduation candidates
auto-skill graduate

Confidence Lifecycle

External Skill (50%)
    │
    ├── 3+ uses, 70% success → Proven (75%)
    │
    ├── 5+ uses, 80% success → Graduated (85%+)
    │                           Local SKILL.md created
    │
    └── Low usage / failures → Confidence decays
                                Eventually dropped

Publishing

If you've built skills that could help others, Auto-Skill can publish them to Skills.sh:

# Find publishable skills (≥85% confidence, ≥5 uses, ≥80% success)
auto-skill discover --publishable

# Sync community stats (installs, ratings)
auto-skill stats --sync

Published skills track community install counts and ratings, so you can see how your workflows help others.