/history

⚡ EXECUTION INSTRUCTIONS FOR CLAUDE

When this command is invoked, YOU (Claude) must execute these steps immediately: This is NOT documentation - these are COMMANDS to execute right now. Use TodoWrite to track progress through multi-phase workflows.

🚨 EXECUTION WORKFLOW

Phase 0: Claude Auto-Memory Read

Action Steps:

1. Discover memory files: Use Python to find all project memory files:

   import glob, os, subprocess
   try:
       git_root = subprocess.check_output(['git', 'rev-parse', '--show-toplevel'], text=True).strip()
       project_key = git_root.replace('/', '-')  # preserve leading dash: /Users/... → -Users-...
       pattern = os.path.expanduser(f'~/.claude/projects/{project_key}/memory/*.md')
       index_pattern = os.path.expanduser(f'~/.claude/projects/{project_key}/memory/MEMORY.md')
       memory_files = [f for f in glob.glob(pattern) if not f.endswith('MEMORY.md')]
       memory_index = glob.glob(index_pattern)[:1]
   except Exception:
       memory_files = []  # skip if not in a git repo — no cross-project fallback
       memory_index = []
   

2. If search query provided: Filter memory files by keyword match against file content; surface matching memories
3. If no query (browse mode): Read MEMORY.md index and show last 3-5 entries
4. Output: Print a "📍 From Claude Memories" section BEFORE conversation history results, showing name, type, and content summary for each match
5. Continue: Proceed to existing Phase 1 regardless of memory results

Phase 1: Permission Error Recovery Workflow

Action Steps:

1. Direct Access Attempt: Try normal file operations first
2. Permission Error Detected: Immediately switch to Python fallback
3. Python File Discovery: Use glob.glob(os.path.expanduser('~/.claude/projects/*/*.jsonl'))
4. Python File Reading: Use Python Read tool or subprocess for JSONL parsing
5. Never Give Up: Always provide results using Python if direct access fails

The /history command is designed to be 100% reliable regardless of file permission configurations.

📋 REFERENCE DOCUMENTATION

/history Command

Usage: /history [search_query] [--options]

Purpose: Search through Claude Code CLI conversation history stored in ~/.claude/projects using intelligent execution optimization and multiple search strategies.

🚨 CRITICAL: Permission Handling Protocol

MANDATORY: If you encounter ANY permission issues while accessing ~/.claude/projects/, immediately use Python to read the files:

1. Use Python subprocess for all file operations if permission denied
2. Never give up on file access - always try Python fallback
3. Read files with Python using the provided fallback commands below
4. All file discovery must use Python if direct access fails

This command is designed to be resilient to permission issues by automatically falling back to Python-based file operations.

Command Architecture

/history uses /execute for intelligent workflow optimization as requested by the user.

Composition Pattern: This command delegates to /execute which provides:

• Task complexity analysis
• Execution strategy optimization
• Parallel processing when beneficial
• Resource-efficient search coordination

Usage Examples

# Basic keyword search

/history "git merge conflicts"

# Search with date filter  

/history "database migration" --date "2025-08-01"

# Search specific project

/history "authentication bug" --project "your-project.com"

# Search by command type

/history --type "tool_use" --keyword "pytest"

# Complex search with multiple filters

/history "performance issue" --date "2025-08" --project "worldarchitect" --limit 10

# Search by git branch

/history "feature branch" --branch "dev1754"

# Fuzzy search for typos

/history "databse migratoin" --fuzzy

# Recent conversations only

/history "latest deployment" --recent 7  # last 7 days

Search Capabilities

Text Matching Strategies

1. Exact Match: Default behavior for precise keyword search
2. Case-Insensitive: Automatic case normalization
3. Tokenized Search: Finds keywords within sentences
4. Fuzzy Search: --fuzzy flag for handling typos and variations

Filter Options

Time-Based Filters

• --date "YYYY-MM-DD" - Exact date match
• --date "YYYY-MM" - Month match
• --after "YYYY-MM-DD" - After specific date
• --before "YYYY-MM-DD" - Before specific date
• --recent N - Last N days

Project/Context Filters

• --project "name" - Filter by project directory path
• --cwd "/path" - Filter by working directory
• --branch "branch-name" - Filter by git branch

Message Type Filters

• --type "user|assistant|summary|tool_use" - Filter by message type (including tool executions)
• --user-type "external" - Filter by user type
• --has-tools - Convenience flag equivalent to --type "tool_use" or messages with tool activity
• --has-errors - Messages with error indicators

Output Control

• --limit N - Maximum results (default: 20)
• --context N - Lines of context around matches (default: 2)
• --format "json|text|table" - Output format (default: text)
• --files-only - Show only matching file names

Implementation Details

Performance Optimization (Based on Research)

1. Streaming JSONL Processing: Line-by-line parsing to avoid memory issues
2. Pre-filtering Pipeline: Filter by high-selectivity metadata first (date, project, session)
3. Parallel File Processing: Process multiple conversation files concurrently
4. Smart File Discovery: Efficiently locate candidate files by directory structure

Search Algorithm

1. Parse search query and extract filters
2. Discover candidate files based on project/date filters
3. For each file (in parallel):
   - Stream parse JSONL line by line
   - Pre-filter by metadata (timestamp, cwd, branch)
   - Apply text search to message content
   - Collect matches with context
4. Aggregate, rank, and format results
5. Apply output limits and formatting

File Format Understanding

Based on research of ~/.claude/projects structure:

Directory Structure:

~/.claude/projects/
├── -home-$USER-projects-worldarchitect-ai/
│   ├── uuid1.jsonl
│   ├── uuid2.jsonl
│   └── ...
└── -home-$USER-projects-other-project/
    └── ...

Note on naming: project directories are sanitized absolute paths where / is replaced with - (e.g., /home/$USER/projects/your-project.com → -home-$USER-projects-worldarchitect-ai). On Windows, backslashes \ are normalized similarly.

JSONL Format (per conversation file):

{"type":"summary","summary":"Conversation summary","leafUuid":"uuid"}
{"parentUuid":null,"userType":"external","cwd":"/path","sessionId":"uuid","version":"1.0.67","gitBranch":"main","type":"user","message":{"role":"user","content":"user message"},"uuid":"uuid","timestamp":"2025-08-09T..."}
{"parentUuid":"uuid","type":"assistant","message":{"role":"assistant","content":"assistant response"},"uuid":"uuid","timestamp":"2025-08-09T..."}

Detection rules:

• Tool usage: records with type:"tool_use" or message.toolCalls/toolUse present.
• Errors: error:true at top-level, or message.level in {"error","fatal"}, or message.content containing standardized error markers. Document any repository-specific variations here to keep the filters accurate.

Integration with /execute

The /history command delegates its execution to /execute with a comprehensive task description that includes secure file access and robust permission handling:

/execute Search through Claude Code CLI conversation history with robust file access:
- Query: [user_search_query]
- Filters: [parsed_filter_options] 
- Strategy: Python-first file access with permission error handling
- Primary Access: Use Python subprocess with explicit file operations
- Fallback Strategy: If permission denied or file access fails, ALWAYS use Python3 directly
- File Discovery: python3 -c "import glob, os; print('\n'.join(glob.glob(os.path.expanduser('~/.claude/projects/*/*.jsonl'))))"
- Content Reading: Use Python Read tool or Python subprocess to read JSONL files
- Permission Recovery: On ANY permission error, immediately switch to Python-based file operations
- Error Handling: Explicit fallback instructions - "Use Python to read files if you encounter permission issues"
- Algorithm: Python file discovery → JSONL parsing → Text search → Context extraction → Results formatting
- Output: Formatted search results with conversation context

CRITICAL PERMISSION HANDLING PROTOCOL: If Claude encounters ANY of the following issues:

1. Permission Denied: Use Python subprocess to read files directly
2. File Access Errors: Switch to Python-based file operations immediately
3. Directory Access Issues: Use python3 -c "import os; os.listdir(os.path.expanduser('~/.claude/projects'))"
4. JSONL Parse Errors: Use Python json module for robust parsing

Python Fallback Commands:

# Directory listing

python3 -c "import os; print('\n'.join(os.listdir(os.path.expanduser('~/.claude/projects'))))"

# File discovery  

python3 -c "import glob, os; print('\n'.join(glob.glob(os.path.expanduser('~/.claude/projects/*/*.jsonl'))))"

# File content reading (single line executable)

python3 -c "import json,sys; [print(json.dumps(json.loads(l)) if l.strip() else '') for l in open(sys.argv[1])]" ~/.claude/projects/-home-user-projects-worldarchitect-ai/uuid.jsonl

# File content reading (with error handling)  

python3 -c "import json,sys; exec('for l in open(sys.argv[1]):\\n  try:\\n    print(json.dumps(json.loads(l)))\\n  except: print(f\"Parse error: {l.strip()}\", file=sys.stderr)')" ~/.claude/projects/-home-user-projects-worldarchitect-ai/uuid.jsonl

This allows /execute to:

• Analyze the search complexity (number of files, query complexity)
• Choose optimal execution strategy (parallel vs sequential)
• Handle resource management and error recovery with Python fallback
• Provide intelligent progress reporting
• Automatically use Python for file operations if permission issues occur

Output Format

Default Text Format

=== Conversation History Search Results ===
Query: "git merge conflicts"
Found: 3 matches in 2 conversations

📁 your-project.com | 📅 2025-08-01 | 🌿 dev1754132318
💬 "I need help resolving git merge conflicts in the authentication module..."
   Context: Discussion about merge strategy and conflict resolution tools
   File: 7859c820-1ba2-41c5-9db1-22c6814455e6.jsonl

📁 your-project.com | 📅 2025-07-28 | 🌿 main  
💬 "The git merge created conflicts in package.json..."
   Context: Package dependency conflicts during feature merge
   File: a1b2c3d4-e5f6-7890-abcd-ef1234567890.jsonl

JSON Format (--format json)

{
  "query": "git merge conflicts",
  "total_matches": 3,
  "total_files": 2,
  "results": [
    {
      "conversation_id": "7859c820-1ba2-41c5-9db1-22c6814455e6",
      "project": "your-project.com", 
      "date": "2025-08-01",
      "branch": "dev1754132318",
      "match": {
        "message": "I need help resolving git merge conflicts...",
        "context": ["Previous message...", "Matched message", "Next message..."],
        "timestamp": "2025-08-01T11:00:11.845Z"
      }
    }
  ]
}

Table Format (--format table)

Query: git merge conflicts | Matches: 3 | Files: 2
┌──────────────────────┬───────────────┬──────────────┬───────────────┬──────────────────────────────────────┐
│ conversation_id      │ project       │ date         │ branch        │ message_snippet                      │
├──────────────────────┼───────────────┼──────────────┼───────────────┼──────────────────────────────────────┤
│ 7859c820-...-55e6    │ worldarchitect│ 2025-08-01   │ dev1754132318 │ I need help resolving git merge ...  │
│ a1b2c3d4-...-7890    │ worldarchitect│ 2025-07-28   │ main          │ The git merge created conflicts ...  │
└──────────────────────┴───────────────┴──────────────┴───────────────┴──────────────────────────────────────┘

Error Handling

• Missing Directory: Graceful handling if ~/.claude/projects doesn't exist - use Python to check with os.path.exists()
• Corrupted JSONL: Skip malformed lines with warning using Python json module
• Permission Issues: AUTOMATIC PYTHON FALLBACK - never fail due to permissions, always try Python subprocess
• Large Result Sets: Automatic pagination and memory management
• No Matches: Helpful suggestions for refining search query
• File Access Failures: Immediate switch to Python-based file operations with detailed error recovery

Integration Features

Secure File Access

The command uses secure direct access to ~/.claude/projects/ directory, avoiding symlink vulnerabilities while maintaining search functionality within the project workspace.

Memory MCP Integration

Search patterns and frequently accessed conversations can be cached in Memory MCP for faster repeated searches.

Command Composition

Can be combined with other commands:

/history "database issue" | /learn    # Learn from historical database solutions
/history "deployment error" | /debug  # Debug with historical context

Advanced Features

Semantic Search (Future Enhancement)

• Leverage embeddings for conceptual similarity matching
• Find conversations about similar topics even with different keywords

Query Intelligence

• Auto-suggest corrections for common typos
• Recommend related search terms based on conversation patterns
• Historical query analysis for better search suggestions

Examples of Complex Searches

# Find all tool usage in August 2025

/history --type "tool_use" --date "2025-08" --format table

# Search for authentication-related conversations in specific project

/history "auth login password" --project "worldarchitect" --limit 5

# Find recent error discussions 

/history "error exception failed" --recent 14 --has-tools

# Fuzzy search with context

/history "databse migratoin problm" --fuzzy --context 5

# Find conversations about specific files

/history "main.py" --project "worldarchitect" --format json

This command provides comprehensive conversation history search capabilities while leveraging /execute for intelligent performance optimization as requested.