/agent-configuration

Agent Configuration Details

Task Tool Invocation

Required Parameters

Task tool parameters:
  subagent_type: "refactor:code-simplifier"
  description: "Refactor code with aggressive cleanup"
  prompt: <detailed context below>

Prompt Context Structure

Pass the following information to the agent:

Refactor the following files with aggressive cleanup enabled:

Target Scope: <list of file paths>

Scope Determination: <one of: "explicit paths", "semantic search: <query>", "session context">

Aggressive Mode: Remove legacy code, unused exports, backwards-compatibility hacks, and rename improperly named variables.

<if semantic search>
Semantic Query: "<original query>"
Search Results: <number> files matched
</if>

<if session context>
Recent Changes: <number> files modified in git working tree
</if>

Example Invocations

Path-based:

Target Scope:
- src/auth/login.ts
- src/auth/session.ts

Scope Determination: explicit paths

Aggressive Mode: Remove legacy code, unused exports, backwards-compatibility hacks, and rename improperly named variables.

Semantic search:

Target Scope:
- src/services/user-validator.ts
- src/lib/validation-utils.ts
- src/middleware/auth-check.ts

Scope Determination: semantic search: "user validation logic"

Semantic Query: "user validation logic"
Search Results: 3 files matched

Aggressive Mode: Remove legacy code, unused exports, backwards-compatibility hacks, and rename improperly named variables.

Session context:

Target Scope:
- src/components/Button.tsx
- src/styles/theme.ts

Scope Determination: session context

Recent Changes: 2 files modified in git working tree

Aggressive Mode: Remove legacy code, unused exports, backwards-compatibility hacks, and rename improperly named variables.

Agent Internal Workflow

The code-simplifier agent executes the following workflow autonomously:

1. Load Best Practices Skill

Load refactor:best-practices skill using Skill tool

This provides:

• Language-specific refactoring rules (TypeScript, Python, Go, Swift)
• Framework detection patterns (Next.js, React)
• Universal code quality principles
• Performance optimization patterns

2. Analyze Target Files

For each file in target scope:

1. Detect Language - Based on file extension
2. Detect Framework - Scan for framework imports (Next.js, React, etc.)
3. Load Relevant References - From best-practices skill based on detection
4. Identify Issues - Complexity, redundancy, code quality violations

3. Apply Refactoring

Aggressive Mode Changes:

• Remove all unused imports, exports, variables, functions
• Delete backwards-compatibility shims (e.g., _unusedVar, re-exports of deleted code)
• Rename misleading identifiers to descriptive names
• Remove commented-out code blocks
• Delete try-catch in trusted code paths
• Eliminate defensive null checks when type system guarantees non-null

Behavior-Preserving Changes:

• Simplify nested ternaries to if-else or guard clauses
• Extract complex conditions to named boolean variables
• Replace magic numbers with named constants
• Modernize syntax (e.g., arrow functions, destructuring)
• Apply framework-specific optimizations when applicable

4. Validate Changes

• Suggest running tests relevant to refactored files
• Provide specific test commands (e.g., npm test auth.test.ts)
• Note manual verification steps for UI changes

5. Generate Summary

Track all changes by category for summary report (see output-requirements.md).

Best Practices Skill Integration

Automatic Loading

The agent's skills manifest includes refactor:best-practices, so it loads automatically on launch.

Framework Detection Examples

Next.js Detection:

import { useRouter } from 'next/router'
import Image from 'next/image'

React Detection:

import React from 'react'
import { useState } from 'react'

Rule Filtering

Agent applies only rules relevant to detected frameworks:

• Next.js rules → only if Next.js detected
• React rules → only if React detected
• Universal rules → apply to all code

Reference File Loading

Based on detection, agent loads:

• TypeScript: references/typescript.md
• Python: references/python.md + references/python/INDEX.md
• Next.js: references/react/ directory with rule index
• Universal: references/universal.md

Model and Performance

Model Configuration

model: opus  # High reasoning for complex refactoring decisions

Opus provides superior:

• Code quality judgment
• Complex pattern recognition
• Multi-file consistency
• Behavioral preservation verification

Allowed Tools

tools:
  - Read       # Read target files
  - Edit       # Apply refactoring changes
  - Glob       # Discover related files
  - Grep       # Search for patterns
  - Bash(git:*)  # Git status, diff for context
  - Skill      # Load best-practices

No Write tool: Agent uses Edit for surgical changes, not full rewrites.

Restricted Bash: Only git commands allowed for safety.