Bundled Skills Reference

Verified beginner beginner 参考型 Reference ⚡ Claude Code 专属 ⚡ Claude Code Optimized
4 min read · 189 lines

5 bundled skills: /batch parallel changes + /simplify quality review + /loop scheduled tasks + /debug troubleshooting

Bundled Skills Reference

Overview

Claude Code ships with 5 bundled skills covering the most common workflows. They work out of the box — no installation required.

/batch — Large-Scale Parallel Changes

Purpose

Orchestrate large-scale parallel changes across a codebase. Breaks work into 5-30 independent units, each running in an isolated git worktree.

Usage

/batch Convert all React class components to function components
/batch Add input validation to every endpoint under src/api/

How It Works

  1. Claude analyzes codebase, identifies files/modules needing changes
  2. Breaks work into independent processing units
  3. Creates a separate git worktree for each unit
  4. Multiple subagents execute changes in parallel
  5. Merges results back to main branch

Good For

  • Cross-file renames/refactors
  • Batch adding type annotations
  • Unifying code style
  • API version migrations
  • Bulk configuration updates

Notes

  • Each unit should be independent (no cross-dependencies)
  • Sweet spot: 5-30 units. Too few isn't worth parallelizing, too many creates overhead
  • Requires a git repository

/simplify — Code Quality Review

Purpose

Review recently changed files for code reuse opportunities, quality issues, and efficiency improvements. Spawns 3 parallel review agents.

Usage

/simplify
/simplify focus on performance optimization

How It Works

  1. Identifies recently modified files
  2. Launches 3 parallel review agents:
    • Reuse review: Finds duplicate code and extractable patterns
    • Quality review: Checks naming, structure, error handling
    • Efficiency review: Spots performance issues and unnecessary complexity
  3. Aggregates report with specific improvement suggestions

Good For

  • Self-check after feature development
  • Quality gate before PR submission
  • Confirming no leftover issues after refactoring

/loop — Scheduled Recurring Tasks

Purpose

Run a prompt or skill repeatedly on a set interval. Ideal for polling, monitoring, periodic checks.

Usage

/loop 5m Check deployment status and report
/loop 2h /simplify
/loop 30s Run npm test and report failing tests

Interval Syntax

  • 30s — 30 seconds
  • 5m — 5 minutes (default: 10 minutes)
  • 2h — 2 hours

Notes

  • Session-scoped, auto-expires after 3 days
  • Up to 50 scheduled tasks per session
  • Also supports natural language one-shot reminders

/debug — Session Debugging

Purpose

Troubleshoot current session issues. Reads the debug log and analyzes error causes.

Usage

/debug
/debug Why did the last command fail

How It Works

  1. Reads current session debug log (transcript.jsonl)
  2. Analyzes recent tool calls and errors
  3. Provides diagnosis and suggestions

Good For

  • Tool call failures with unclear causes
  • Claude behaving unexpectedly
  • Understanding what happened in a session

/claude-api — API Reference

Purpose

Loads Claude API reference material for your project's language, plus Agent SDK reference.

Usage

/claude-api

How It Works

  1. Detects project's programming language
  2. Loads corresponding Claude API documentation
  3. Loads Agent SDK reference (if applicable)
  4. Provides code examples and best practices

Good For

  • Integrating Claude API into your application
  • Using Anthropic SDKs (Python/TypeScript)
  • Building custom agents with Agent SDK

GitHub Example Commands

Beyond the 5 bundled skills, the Claude Code repo provides 3 example commands:

Command File Function
commit-push-pr .claude/commands/commit-push-pr.md Commit, push, and create a PR
dedupe .claude/commands/dedupe.md Find and eliminate duplicate code
triage-issue .claude/commands/triage-issue.md Analyze and triage GitHub Issues

These serve as useful reference templates for creating custom skills.

相关技能 Related Skills