Find Development Tasks

I'll locate all TODO comments and unfinished work markers in your codebase.

Token Optimization

Target Reduction: 70% (2,000-3,000 → 600-900 tokens)

Optimization Status: ✅ Optimized (Phase 2 Batch 3D-F, 2026-01-26)

Key Optimizations

1. Pure Grep-Based Search (No File Reading)

   • Use Grep tool exclusively for TODO/FIXME/HACK/XXX/NOTE detection
   • Pattern: "TODO|FIXME|HACK|XXX|NOTE" with case-insensitive search
   • Grep provides file location, line numbers, and surrounding context
   • NEVER use Read tool - grep output contains all necessary information
   • Saves: 1,000-1,500 tokens per search

2. Git Diff Scoping (Smart Defaults)

   • Default behavior: Search only files changed in current branch
   • Use git diff --name-only main...HEAD to get changed files
   • Pass file list to Grep tool with path restriction
   • Fall back to full codebase search only if no git repo or explicit request
   • Saves: 500-1,000 tokens for large codebases

3. Head Limit on Results

   • Limit grep output to first 50 TODO comments using head_limit: 50
   • Prevents overwhelming results in large codebases
   • Most actionable TODOs are in recently changed files anyway
   • User can request "show all" if needed for comprehensive audit
   • Saves: 500-1,000 tokens for codebases with many TODOs

4. Early Exit Detection

   • Check if grep finds any TODOs before processing results
   • If no TODOs found: Immediately report "No TODOs found" and exit
   • Avoid creating empty task lists or unnecessary analysis
   • Saves: 200-400 tokens when no TODOs present

5. Progressive Disclosure (Priority-Based)

   • Sort results by marker type: FIXME/HACK/XXX first, then TODO, then NOTE
   • Show critical items (FIXME/HACK) in detail with context
   • Summarize lower-priority items (TODO/NOTE) concisely
   • Ask user if they want full details on non-critical items
   • Saves: 300-500 tokens by focusing on high-priority items

6. Result Caching

   • Cache TODO inventory at .claude/cache/find-todos/todo-inventory.json
   • Format: {"file": "path", "line": 42, "type": "TODO", "comment": "text", "context": "code"}
   • Subsequent skills (fix-todos, todos-to-issues) reuse cached data
   • Cache expires after 1 hour or when git HEAD changes
   • Saves: 500-1,000 tokens for dependent skills

Usage Patterns

Minimal scope (200-300 tokens):

find-todos path/to/specific/file.ts

Changed files only (400-600 tokens):

find-todos

Full codebase audit (600-900 tokens):

find-todos --all

Expected Token Usage

Scenario	Before	After	Savings
Specific file	800-1,000	200-300	70%
Changed files (default)	1,500-2,000	400-600	70%
Full codebase	2,500-3,000	600-900	70%
No TODOs found	600-800	100-200	75%

Caching Behavior

• Cache location: .claude/cache/find-todos/
• Cached data: TODO inventory by file with line numbers and context
• Shared with: /fix-todos, /create-todos, /todos-to-issues skills
• Cache invalidation: 1 hour timeout or git HEAD change
• Cache benefits: Enables zero-cost TODO lookups for dependent skills

I'll use the Grep tool to efficiently search for task markers with context:

• Pattern: "TODO|FIXME|HACK|XXX|NOTE"
• Case insensitive search across all source files
• Show surrounding lines for better understanding

For each marker found, I'll show:

1. File location with line number
2. The full comment with context
3. Surrounding code to understand what needs to be done
4. Priority assessment based on the marker type

When I find multiple items, I'll create a todo list to organize them by priority:

• Critical (FIXME, HACK, XXX): Issues that could cause problems
• Important (TODO): Features or improvements needed
• Informational (NOTE): Context that might need attention

I'll also identify:

• TODOs that reference missing implementations
• Placeholder code that needs replacement
• Incomplete error handling
• Stubbed functions awaiting implementation

After scanning, I'll ask: "Convert these to GitHub issues?"

• Yes: I'll create properly categorized issues
• Todos only: I'll maintain the local todo list
• Summary: I'll provide organized report

Important: I will NEVER:

• Add "Created by Claude" or any AI attribution to issues
• Include "Generated with Claude Code" in descriptions
• Modify repository settings or permissions
• Add any AI/assistant signatures or watermarks

This helps track and prioritize unfinished work systematically.