What Are Skills?
Skills are reusable procedures and domain knowledge that can be invoked by commands, agents, or users. They provide structured workflows for common coding tasks, enforce best practices, and capture learnings that compound over time.11 Production Skills
Battle-tested workflows from real development sessions
Cross-Agent Support
Works with Claude Code, Cursor, and 32+ agents via SkillKit
Learning System
Self-correcting patterns that improve with every session
Orchestration Ready
Wire skills into multi-phase workflows with validation gates
Core Skills
The Pro Workflow system includes 11 specialized skills:Pro Workflow - Complete AI Coding System
Pro Workflow - Complete AI Coding System
The master skill containing all core patterns:
- Self-correction loop
- Parallel worktrees for zero dead time
- Context discipline and token management
- Learning capture and replay
- Multi-phase development orchestration
Smart Commit - Quality Gates & Commits
Smart Commit - Quality Gates & Commits
Automate quality checks and create well-crafted commits:
- Run lint, typecheck, tests on affected files
- Scan for console.log, TODOs, secrets
- Generate conventional commit messages
- Stage specific files (never
git add .)
Wrap-Up - End-of-Session Ritual
Wrap-Up - End-of-Session Ritual
End sessions with intention:
- Audit modified files and uncommitted changes
- Run quality checks
- Capture learnings from mistakes
- Generate session summary
Learn Rule - Capture Corrections
Learn Rule - Capture Corrections
Turn mistakes into permanent memory:
- Extract lessons from corrections
- Categorize by type (Navigation, Testing, Git, etc.)
- Store with full context
- Surface in future sessions
Parallel Worktrees - Zero Dead Time
Parallel Worktrees - Zero Dead Time
Work on multiple tasks simultaneously:
- Create isolated git worktrees
- Run parallel sessions while tests run
- Manage background agents
- Clean up automatically
Replay Learnings - Apply Past Lessons
Replay Learnings - Apply Past Lessons
Surface relevant learnings before starting work:
- Search past corrections by keyword
- Rank by relevance to current task
- Flag high correction-rate areas
- Suggest approaches based on history
Session Handoff - Resume Context
Session Handoff - Resume Context
Generate handoff documents for session continuity:
- Current branch and commit status
- Completed, in-progress, pending tasks
- Key decisions and reasoning
- Copy-pasteable resume command
Insights - Session Analytics
Insights - Session Analytics
Track patterns and productivity:
- Learning category breakdown
- Correction rate trends
- Most/least applied patterns
- Stale learnings that need review
Deslop - Remove AI Code Bloat
Deslop - Remove AI Code Bloat
Clean up AI-generated over-engineering:
- Remove unnecessary comments
- Strip defensive try/catch blocks
- Simplify premature abstractions
- Delete backwards-compatibility hacks
Context Optimizer - Token Management
Context Optimizer - Token Management
Manage context window and token budget:
- Proactive compaction at 50%
- MCP audit (keep <10 MCPs, <80 tools)
- Subagent delegation patterns
- CLAUDE.md optimization
Orchestrate - Multi-Phase Development
Orchestrate - Multi-Phase Development
Wire Commands, Agents, and Skills together:
- Research → Plan → Implement → Review
- Validation gates between phases
- Confidence scoring (GO/HOLD)
- Never proceed without approval
How Skills Work
Two Types of Skills
Agent skills use context tokens but are always available. On-demand skills load only when called, keeping main context clean.
Invocation Methods
| Method | Example | When to Use |
|---|---|---|
| Command | /smart-commit | User-triggered workflows |
| Agent preload | skills: ["pro-workflow"] | Domain knowledge always needed |
| Skill tool | Skill("replay-learnings") | AI-initiated, conditional use |
Usage Patterns
Pattern 1: Self-Correction Loop
The most powerful pattern - Claude learns from corrections automatically:Pattern 2: Session Lifecycle
Structured workflow from start to finish:Pattern 3: Multi-Phase Development
For complex features requiring validation gates:Installation
- Cursor
- Claude Code
- Manual
- Any Agent (SkillKit)
Quick Start
Minimal Setup (30 seconds)
Add to your CLAUDE.md:First Session
- Start with context:
/replay what do I know about auth - Make changes: Claude implements with quality gates
- Capture mistakes: “Remember this” triggers learn-rule
- Commit:
/smart-commitruns quality checks automatically - End intentionally:
/wrap-upcaptures learnings and generates summary - Handoff:
/handoffcreates resume context for next session
Integration with Pro Workflow
Skills work seamlessly with other Pro Workflow components:Agents
5 specialized agents with skills preloaded
Commands
10 slash commands that invoke skills
Hooks
18 hook events that trigger skills
Rules
6 Cursor rules enforce skill patterns
Philosophy
Compound Improvements
Small corrections lead to big gains over time. After 50 sessions, Claude barely needs correcting.
Trust but Verify
Let AI work between checkpoints. Review at planning, milestones, destructive operations.
Memory is Precious
Both yours and Claude’s. Context discipline, proactive compaction, subagent delegation.
Next Steps
Learn the Core Workflow
Master the complete AI coding workflow system
Start with Smart Commit
Add quality gates to your commit process
Try Multi-Phase Dev
Build features through structured phases
Optimize Context
Manage your token budget effectively