Type: Bundled skill (compiled into Claude Code binary)
Invocation:
/simplifyor/simplify focus on [area]The
/simplifycommand is a built-in skill that performs an automated, multi-agent code review and cleanup pass on recently changed files. It is designed to be run after implementing a feature or fixing a bug, before opening a pull request.
This is a bundled skill, meaning it ships inside the Claude Code binary and does not require installation. Unlike custom skills placed in .claude/skills/, this skill is maintained by Anthropic and updated with Claude Code releases. The exact SKILL.md source is compiled into the binary and not publicly accessible.
The following documentation is reconstructed from public documentation, community sources, and behavioral analysis.
The skill identifies the scope of review by inspecting recent changes:
- If inside a Git repository, runs
git diff(orgit diff HEADfor staged changes) to determine what was modified - If no Git repository is detected, reviews the most recently modified files
- The diff output defines the exact scope of the review
Three specialized sub-agents are spawned in parallel, each receiving the full diff for context:
Reviews changes to identify:
- Duplicated logic that could be extracted into shared functions
- Existing utility functions or helpers that should be used instead of new code
- Redundant code blocks and repeated patterns
- Opportunities to refactor into reusable components or modules
Evaluates code structure and style:
- Naming consistency and readability
- Function decomposition and control flow clarity
- Compliance with coding standards defined in
CLAUDE.md - Code smells such as leaky abstractions, stringly-typed code, unnecessary nesting
- Over-engineering, unnecessary abstractions, or premature optimization ("gold-plating")
Analyzes performance and resource usage:
- Unnecessary allocations and redundant computations
- Loops that could be batched or optimized
- N+1 query patterns and inefficient file or network access
- Missed concurrency opportunities
- Unnecessary re-renders (in frontend code)
Once all three agents complete their reviews:
- Findings are aggregated and deduplicated
- False positives or findings not worth addressing are skipped
- Fixes are applied directly to the codebase
- A summary of what was changed is reported to the user
/simplify # Review all recent changes
/simplify focus on memory efficiency # Focus on specific concern
/simplify focus on performance # Target performance issues
The effectiveness of /simplify depends heavily on the project's CLAUDE.md file. The quality agent uses coding standards and preferences defined there to align its suggestions with the team's conventions. Projects with well-defined CLAUDE.md files get significantly better results.
| Property | Detail |
|---|---|
| Agent count | 3 parallel sub-agents |
| Scope | Recent changes only (via git diff) |
| Execution | Agents run in parallel for efficiency |
| Fix mode | Automatic application with summary |
| False positives | Skipped rather than argued |
| Customization | Accepts focus areas as arguments |
| Dependency | Reads CLAUDE.md for project-specific standards |
The recommended workflow is:
Build feature → Run tests → /simplify → Review diffs → Commit
This positions /simplify as a post-implementation cleanup pass that tightens AI-generated code, which tends toward verbosity. It is not a replacement for human code review but automates the mechanical cleanup portion.
| Command | Relationship |
|---|---|
/init |
Sets up CLAUDE.md that /simplify reads for standards |
/diff |
Preview changes before running /simplify |
/batch |
For large-scale refactors across entire codebases (alternative to /simplify which targets recent changes only) |