export

Export a session transcript as Markdown or JSON across Claude Code, Codex, GitHub Copilot (CLI and VS Code), Pi, and OpenClaw.

Usage

claudex export <selector> [--format markdown|json]
                           [-o/--output <path>]
                           [-p/--project <substr>]

Flags

Flag	Default	Description
<selector>	—	Session ID prefix or project name substring. Required.
--format	markdown	markdown or json.
-o, --output <path>	stdout	Write to a file instead of stdout.
-p, --project <substr>	—	Disambiguate when the selector matches more than one session.

Example

# List recent sessions to find a prefix
claudex sessions --limit 5

# Export as Markdown to stdout
claudex export e1a2f4

# Save to a file
claudex export e1a2f4 --output session.md

# JSON instead
claudex export e1a2f4 --format json --output session.json

# Ambiguous prefix? Add --project
claudex export ab --project claudex

Markdown output

The Markdown export renders the full conversation as:

# Session: <short-id>

**Project:** <path>
**Date:** <YYYY-MM-DD HH:MM UTC>
**Model:** <model>

---

## User

_<timestamp>_

<content>

## Assistant

_<timestamp>_

<content>

Each turn is a section. Tool calls and tool results are inlined. This is the shape you want for pasting into docs, wikis, or PR descriptions.

JSON output

If the selector resolves to one session, JSON export is one object. If it resolves to multiple sessions, JSON export is an array of those objects. Each object has top-level keys: session_id, project, date, model, message_count, messages, normalized_messages, and records. messages and records preserve every raw JSONL record from the source session file, in chronological order; normalized_messages is the cross-provider role/text view.

{
  "session_id": "<uuid>",
  "project": "<path>",
  "date": "<iso-8601>",
  "model": "<model>",
  "message_count": 14,
  "provider": "claude",
  "file_path": "<absolute-path>",
  "extras": null,
  "normalized_messages": [
    { "role": "user", "timestamp": "<iso-8601>", "text": "..." }
  ],
  "messages": [
    {
      "type": "user",
      "uuid": "...",
      "parentUuid": null,
      "sessionId": "<uuid>",
      "timestamp": "<iso-8601>",
      "cwd": "<path>",
      "gitBranch": "<branch>",
      "version": "<claude-code-version>",
      "userType": "external",
      "isSidechain": false,
      "permissionMode": "default",
      "entrypoint": "cli",
      "promptId": "...",
      "message": { "role": "user", "content": "..." }
    }
  ],
  "records": [
    {
      "type": "user",
      "uuid": "...",
      "parentUuid": null,
      "sessionId": "<uuid>",
      "timestamp": "<iso-8601>",
      "cwd": "<path>",
      "gitBranch": "<branch>",
      "version": "<claude-code-version>",
      "userType": "external",
      "isSidechain": false,
      "permissionMode": "default",
      "entrypoint": "cli",
      "promptId": "...",
      "message": { "role": "user", "content": "..." }
    }
  ]
}

For Claude, raw records keep the Anthropic-shaped message payload. Codex, Copilot, Pi, and OpenClaw records keep their native provider shapes (VS Code Copilot Chat exports the replayed session document).

Useful jq:

# Count messages by role
claudex export <sid> --format json \
  | jq '[.normalized_messages[].role] | group_by(.) | map({role: .[0], count: length})'

# Extract only the text content
claudex export <sid> --format json \
  | jq -r '.normalized_messages[] | select(.role == "assistant") | .text'

Notes

• Selector precedence. Session-ID prefix wins when it matches one or more sessions, even when --project is set. If the selector looks like a Claude session-ID prefix and matches nothing, claudex does not fall back to project-name matching.
• No --json flag. Use --format json — it's the format of the export, not a summary layer.
• Multiple matches. Markdown writes the matching sessions sequentially. JSON returns a proper array instead of concatenated objects.
• Large sessions. Markdown export is streaming where possible, but huge sessions (100+ MB JSONL) can produce very long Markdown files. Use --format json if you want to slice the output yourself.