Bundled Skills
The Built-in Power Tools
Claude Code comes with a set of Skills that are ready to use out of the box. These "Bundled Skills" cover common developer workflows and simultaneously demonstrate what's possible with the Skill system.
/batch -- Parallel Bulk Operations
Imagine you need to make a change across 50 files. Doing them one by one would be painfully slow. /batch uses Git Worktrees to distribute the work across multiple parallel agents.
> /batch Refactor all API endpoints from callback-based to async/await
Claude automatically creates:
- A plan of which files are affected
- Parallel worktrees for simultaneous editing
- A summary of changes for review
iGit Worktrees
Worktrees are separate working copies of the same repo. This allows Claude to edit multiple files simultaneously without conflicts. More on this in the "Subagents & Delegation" module.
/simplify -- Code Review by Three Agents
Complexity is the enemy of maintainability. /simplify launches three parallel review agents that analyze your code from different perspectives:
> /simplify src/lib/dataProcessor.ts
The three agents check:
- Structural complexity -- Nesting, dependencies, abstractions
- Readability -- Naming, comments, comprehensibility
- Performance -- Unnecessary computations, optimization potential
Afterwards, Claude consolidates the results and carries out the agreed simplifications.
/loop -- Periodic Repetition
Some tasks need regular execution. /loop repeats a prompt at set intervals:
> /loop 30s Check if the tests are still green
> /loop 5m Summarize the latest Git commits
Useful during longer refactoring sessions when you need continuous feedback.
/debug -- Systematic Debugging
Instead of aimlessly searching for bugs, /debug activates a structured debug workflow:
> /debug The API sporadically returns 500 errors
Claude activates targeted logging, systematically reproduces the problem, and analyzes the logs. Particularly valuable for hard-to-reproduce errors.
/claude-api -- Load API Reference
When working with the Claude API or the Agent SDK, this Skill loads the current reference documentation into the context:
> /claude-api How do I create a streaming client?
Claude then has the complete API documentation available and can give precise, up-to-date answers.
The AskUserQuestion Tool
Some Bundled Skills use a powerful tool: AskUserQuestion. With it, Claude can ask you structured questions -- not as free text, but as choices with descriptions.
How It Works
Claude presents you with 2-4 options per question, each with a brief description. You select one or more (multi-select is possible). You can always choose "Other" and provide your own answer.
Typical use cases:
- Requirements gathering -- "Which features should the component have?"
- Architecture decisions -- "Which state management fits your project?"
- Config wizards -- "How should the deployment pipeline be configured?"
*Effective Interaction
When Claude presents options, read the descriptions carefully. They often contain important trade-offs that help with your decision. "Other" is always an option -- you're never limited to the predefined answers.
Dynamic Context with Shell Injection
Skills can embed shell commands whose output is executed before being passed to Claude. This works with backtick expressions:
PR diff: !`gh pr diff`
Current branch: !`git branch --show-current`
Recent commits: !`git log --oneline -5`
The backtick expressions are resolved before processing by Claude. This way, Claude always has the current state without you having to manually insert information.
!Security Note
Shell injections are executed before being passed to Claude. Make sure you only embed trusted commands -- especially in shared Skills distributed via Git.
Practical Example: Automated PR Review
A Skill that uses dynamic context to review PRs:
---
name: pr-review
description: Review the current pull request
disable-model-invocation: true
---
# PR Review
Analyze this pull request:
## Diff
!`gh pr diff`
## PR Description
!`gh pr view --json title,body --jq '.title + "\n\n" + .body'`
Check for:
1. Breaking changes
2. Missing tests
3. Security issues
4. Performance implications
Here, the current PR diff and description are loaded automatically -- you don't need to copy anything manually.
What does the `/batch` Skill do differently from normal sequential work?