aviz85/search-history

Search Claude Code Conversation History

Search across all Claude Code sessions and projects to find past discussions, solutions, and context.

Usage

/search-history "aviz-museum" /search-history "supabase migration" --project Users-aviz-dreemz-backend /search-history "Riley Brown" --user-only --limit 5

Parameters

• <search-term> -- keyword or phrase to search (required)
• --project <name> -- limit to specific project dir (e.g., -Users-aviz-sb)
• --user-only -- only search user messages (skip assistant/progress/system)
• --limit N -- max sessions to show (default: 10)

How It Works

Storage Structure

~/.claude/projects/
  -Users-aviz-sb/                    # Project dir (path with / -> -)
    ae1972bf-...-2e6f1b34be03.jsonl   # Session file (UUID)
    ae1972bf-.../                      # Session artifacts
      subagents/                      # Subagent logs
      tool-results/                   # Tool output files
  -Users-aviz-dreemz-backend/
    ...

JSONL Message Types

Each line is a JSON object with a type field:

| Type | Description | Key Fields | |------|-------------|------------| | user | User input | message.content (string or array), timestamp, cwd | | assistant | Claude response | message.content (array of text/tool_use blocks), message.model | | progress | Tool execution tracking | data.toolName, data.status | | system | System events | Compaction, boundaries | | file-history-snapshot | File backup metadata | snapshot.timestamp | | queue-operation | Inbox operations | operation | | last-prompt | Last user prompt text | prompt |

Content Extraction

• User messages: message.content is usually a string, sometimes array of content blocks.
• Assistant messages: message.content is array -- extract .text from type: "text" blocks.

Implementation

Search using grep + Python for structured output:

# Quick grep across all projects
grep -rl "SEARCH_TERM" ~/.claude/projects/*/*.jsonl

For structured search with context, use Python to parse JSONL files, extract matching messages, and display with session metadata (project, timestamp, message type).

Tips

• Start broad -- grep is fast across all JSONL files
• Use --user-only to find what YOU asked, not what Claude replied
• Use --project to narrow to a specific workspace
• Session UUIDs can be used with claude --resume SESSION_ID
• Tool results are in separate .txt files under session-dir/tool-results/
• Subagent conversations are in session-dir/subagents/agent-*.jsonl