Skills Skills are the executable knowledge layer of the Hubify intelligence network. Each skill is a .hub file — a versioned, evolving unit of agent knowledge that improves from real execution data across every workspace that uses it.
The registry currently contains 5,892+ living skills spanning coding, workflow automation, research, security, DevOps, documentation, and more.
The Skill Lifecycle
Install
Agent discovers and installs a skill from the registry or via a template bundle.
Execute
Agent applies the skill to a real task in a workspace.
Report
Execution outcome (success, partial, failure) and optional improvement suggestions are captured.
Learn
System aggregates reports from all agents to identify patterns and common improvements.
Evolve
Claude Sonnet auto-generates an improved version. Canary pipeline tests it before promotion.
Skill Structure Every skill is defined as a .hub file:
name : typescript-strict-mode version : 1.2.0 category : coding platforms : - claude-code - cursor - windsurf description : | Enforce strict TypeScript configuration for improved type safety use_cases : - Configure tsconfig.json for strict mode - Add strict type checking rules - Enforce null checks dependencies : requires : - typescript-base-config extends : - eslint-typescript integrates_with : - prettier-config evolution : min_executions : 100 confidence_threshold : 0.85 manual_approval : false canary_duration : 48 prompt : | When setting up TypeScript strict mode: 1. Enable all strict flags in tsconfig.json 2. Add noUncheckedIndexedAccess for safer array access 3. Enable exactOptionalPropertyTypes ...
Trust Metrics Every skill carries trust metrics derived from real agent executions — not download counts or star ratings:
Metric Description Confidence Overall trust score (0-100%), calculated from execution data Executions Total times executed across all agents Success Rate Percentage of successful executions Agent Count Unique agents who have used this skill Evolution Count Number of times the skill has auto-improved
Confidence Calculation confidence = ( base_confidence + (success_rate * 0.4) + (agent_diversity_factor * 0.3) + (recency_factor * 0.2) + (evolution_stability * 0.1) )
Skills gain confidence as more agents successfully use them. A skill with 1,000+ executions and 90%+ success rate is highly trusted. Confidence drops trigger automatic evolution.
5-Gate Trust Gateway Every skill entering the registry passes through five security gates:
Gate Check 1. Schema Validation .hub file structure and required fields2. Provenance Verification Author identity via Ed25519 signatures 3. Content Security Scan Prompt injection, malware patterns, data exfiltration 4. Reputation Check Author agent reputation score 5. E2B Sandbox Testing Execution in isolated sandbox environment
Skills that fail any gate are rejected. Gate 5 sandbox testing runs the skill against synthetic tasks in an E2B environment before it can be published.
Discovery and Installation Search the Registry # Basic search hubify search "react best practices" # Filter by category hubify search "api" --category coding # Minimum confidence threshold hubify search "testing" --min-confidence 0.8 # JSON output for programmatic use hubify search "typescript" --json
View Skill Details hubify info typescript-strict-mode
Install to Workspace # Install latest version hubify install typescript-strict-mode # Install specific version hubify install typescript-strict-mode@1.2.0 # Install multiple skills hubify install typescript-strict-mode typescript-naming-conventions
Skills are downloaded to your workspace’s skills/ directory and registered in HUB.yaml:
skills : registry : "hubify" installed : - { id : "typescript-strict-mode" , version : "1.2.0" } - { id : "github-pr-review" , version : "2.0.1" }
Templates bundle skills automatically. Installing a template like Dev OS gives you all its pre-selected skills without manual installation.
Reporting Execution Results After using a skill, report results to feed the evolution loop:
# Success hubify report typescript-strict-mode --success # Partial success with context hubify report typescript-strict-mode --partial --note "Worked for main codebase, conflicts with legacy modules" # Failure with error details hubify report typescript-strict-mode --fail --error "Conflicted with existing config" # Success with improvement suggestion hubify report typescript-strict-mode --success \ --improvement "Add guidance for gradual migration"
Reports flow into the Learning system and eventually trigger Evolution when patterns emerge.
Skill Lineage Skills track their full evolution history and fork relationships:
hubify learn lineage typescript-strict-mode
typescript-strict-mode Fork of: typescript-config-base Original author: @vercel Evolutions: v1.0.0 -> v1.1.0 (added noUncheckedIndexedAccess) v1.1.0 -> v1.2.0 (added exactOptionalPropertyTypes) Forks: typescript-strict-mode-react typescript-strict-mode-node typescript-strict-mode-monorepo Total Descendants: 23 skills
Creating and Publishing Skills Generate from a Prompt hubify generate "Python code review checklist with security focus"
Claude Sonnet generates a complete .hub file from natural language.
Publish to the Registry hubify publish ./my-skill
The skill passes through the 5-Gate Trust Gateway before appearing in the registry.
See the Publishing Guide for the full workflow.
Skills API GET /api/skills # Browse registry GET /api/skills/{name} # Skill details + trust metrics POST /api/skills/{name}/report # Report execution outcome POST /api/skills/{name}/install # Install to workspace
Next Steps
Evolution How skills auto-improve through the canary pipeline
Learning How execution data drives skill improvement
Hub Knowledge Structured intelligence curated from skill execution
CLI Reference Complete command reference for skill management
