Claude Code Transcript Search: Navigate Long Sessions with "/"

Claude Code

Claude Code v2.1.83 adds in-session transcript search, accessible by pressing "/" in transcript mode (Ctrl+O). Users can type a search query and step through all matching results in the session history using "n" (forward) and "N" (backward), following the same navigation convention as vim, less, and man. The feature addresses a real usability gap for developers running long agentic sessions where finding a specific tool result, decision point, or error message previously required manually scrolling through hundreds of turns.


In-Session Transcript Search Comes to Claude Code

Claude Code v2.1.83 ships transcript search, a quality-of-life improvement that makes long agentic sessions significantly more navigable. The feature is triggered by pressing / inside transcript mode (entered with Ctrl+O) and uses the familiar n/N key bindings from vim, less, and man to step through matches.

The Navigation Problem in Long Sessions

As Claude Code sessions grow to cover dozens or hundreds of turns β€” common in autonomous refactoring runs, long debugging sessions, or background agent workflows β€” navigating transcript history becomes a friction point. Before v2.1.83, finding a specific tool result, a particular error message, or the point where the agent made a key decision required scrolling backward through the entire transcript.

The community had partially addressed this through external tools: claude-history (a Rust CLI with fuzzy-search across all session files), the claude-code-transcripts HTML export tool, and a VS Code extension offering a searchable sidebar. These tools work on session files on disk but cannot search the live session currently running in the terminal.

How It Works

Pressing / while in transcript mode (Ctrl+O) opens a search prompt. As the user types, Claude Code highlights all matching strings in the visible transcript. Pressing n jumps forward to the next match; N jumps backward. The pattern is immediately familiar to any developer who uses vim, less, or grep-based terminal workflows β€” no new keybindings to learn.

The search works across all content visible in the transcript, including tool call inputs and outputs, assistant responses, user messages, and hook output. A developer can search for a specific filename, error string, function name, or any text that appeared anywhere in the session.

Practical Use Cases

  • Debugging long runs: Search for the first occurrence of an error message to find exactly where a tool call went wrong.
  • Audit and review: Find all instances where the agent wrote to a particular file or called a specific tool.
  • Context recovery: After returning to a long session, quickly locate the last relevant decision point without scrolling.
  • Verification: Confirm that a specific action was or was not taken during an autonomous run.