Skip to main content

​
auto-skill discover

Analyze your current project to discover relevant patterns and suggest skills. Auto-Skill scans local workflow patterns and optionally queries external skill sources like Skills.sh.

​
Usage

auto-skill discover [options]

​
Options

​
--project
string
default:"current directory"
Path to the project to analyze. Defaults to the current working directory.
​
--no-mental
boolean
default:"false"
Disable Mental CLI integration for semantic analysis
​
--no-external
boolean
default:"false"
Disable Skills.sh external skill discovery
​
--limit
string
default:"10"
Maximum number of skill suggestions to return
​
--json
boolean
default:"false"
Output results in JSON format

​
How It Works

1

Project Context Analysis

Analyzes files, languages, and frameworks in the specified project directory
2

Local Pattern Detection

Scans your recent workflow history for repeated tool sequences and successful patterns

External Skill Search

Queries Skills.sh for community skills matching your project context

Mental Model Integration

Uses Mental CLI (if installed) for semantic codebase understanding
5

Recommendation Ranking

Scores and ranks suggestions based on relevance and confidence

​
Discovery Methods

​
Local Pattern Detection

Analyzes your workflow history to detect:
  • Repeated sequences - Same tool calls 3+ times
  • Success patterns - High success rate on similar tasks
  • Error recovery - Successful self-correction workflows
  • Design patterns - Architectural patterns (MVC, Factory, TDD, etc.)

​
External Skill Discovery

Searches 27,000+ community skills from:
  • Skills.sh registry - Verified community skills
  • RFC 8615 well-known endpoints - Organization skill catalogs
  • GitHub repositories - Skills with SKILL.md files

​
Hybrid Recommendations

Combines local and external discovery to suggest:
  • Graduation candidates - Local patterns matching proven external skills
  • Enhancement opportunities - External skills that complement your workflow
  • Best practices - Battle-tested patterns from the community

​
Examples

​
Discover for Current Project

auto-skill discover

Discovering skills for: /Users/username/projects/my-app

No suggestions found yet. Work on some files first.
Auto-Skill needs some workflow history before it can detect patterns. Use your coding agent normally, and patterns will appear after 3+ occurrences.

​
Discover for Specific Project

auto-skill discover --project ~/projects/my-app
Analyzes a specific project directory instead of the current directory.

​
Disable External Discovery

auto-skill discover --no-external
Only analyze local patterns without querying Skills.sh or external sources. Useful for:
  • Offline development
  • Private/sensitive projects
  • Faster discovery (local-only)

​
Disable Mental Integration

auto-skill discover --no-mental
Skip semantic codebase analysis via Mental CLI.

​
Limit Results

auto-skill discover --limit 5
Return only the top 5 most relevant suggestions.

​
JSON Output

auto-skill discover --json

{
  "project_path": "/Users/username/projects/my-app",
  "count": 0,
  "suggestions": []
}

​
Pattern Detection Criteria

Patterns are detected when they meet these thresholds (configurable in ~/.claude/auto-skill.local.md):
​
min_occurrences
number
default:"3"
Minimum repetitions to detect a pattern
​
min_confidence
number
default:"0.7"
Minimum confidence score (0.0-1.0) to suggest
​
lookback_days
number
default:"7"
Days of history to analyze

​
Confidence Scoring

Suggestions are scored based on:
ConfidenceSourceMeaning
50%ExternalRelevant based on project context
75%LocalProven pattern in your workflow
85%GraduatedBattle-tested (used 5+ times, 80%+ success)
90%+HybridLocal pattern + matching community skill

Graduate Skills

Promote local patterns to permanent skills

View Telemetry

Check which patterns are most effective

​
Installation

To install discovered skills, use the Skills CLI: 
npx skills add owner/repo

Build docs developers (and LLMs) love

Get started for freeTalk to us