cctrace

Export and import Claude Code sessions.

Claude Code stores conversation history in ~/.claude/projects/<normalized-path>/*.jsonl. This tool extracts those sessions into portable formats for archival, analysis, or sharing.

What's New

v2.0 adds portable sessions: export a session to your git repository, push it, and let someone else import it to continue where you left off. The original export functionality remains unchanged.

Two Modes of Operation

Classic Export

Exports to ~/claude_sessions/exports/ with timestamped directories. This is the original behavior and remains the default.

python3 export_claude_session.py

Produces:

~/claude_sessions/exports/2025-07-02_16-45-00_f33cdb42/
├── raw_messages.jsonl     # Original session data
├── conversation_full.md   # Human-readable markdown
├── conversation_full.xml  # Structured XML with full metadata
├── session_info.json      # Session metadata
└── summary.txt            # Statistics

Portable Export

Exports to .claude-sessions/<name>/ within your repository. Designed to be committed and shared.

python3 export_claude_session.py --in-repo --export-name my-session

Produces:

.claude-sessions/my-session/
├── .cctrace-manifest.json    # Required for import
├── RENDERED.md               # Renders on GitHub
├── session/
│   ├── main.jsonl            # Session transcript
│   ├── file-history/         # File snapshots from undo history
│   ├── todos.json            # Todo state
│   └── plan.md               # Plan file if present
├── config/
│   ├── commands/             # Slash commands
│   ├── skills/               # Skills
│   └── ...                   # Other .claude/ config
└── [legacy files]            # For backwards compatibility

The portable export includes everything needed to resume the session: conversation history, file snapshots, todos, plans, and project-specific Claude Code configuration.

Importing a Session

Someone clones your repo and wants to continue your session:

python3 import_session.py .claude-sessions/my-session/
claude -c  # Lists available sessions, including the imported one

What Import Does

1. Validates the export has a .cctrace-manifest.json
2. Creates a pre-import snapshot for recovery
3. Generates a new session ID (your local sessions remain unaffected)
4. Rewrites internal UUIDs while preserving message threading
5. Updates cwd paths to the local project directory
6. Copies session file to ~/.claude/projects/<normalized-path>/
7. Imports auxiliary data (file-history, todos, plans)
8. Imports config files (commands, skills, hooks, agents, rules)

What Import Preserves

Certain fields are cryptographically signed or tied to Anthropic's API and must not be modified:

• message.id (Anthropic message ID)
• requestId (Anthropic request ID)
• signature in thinking blocks
• tool_use.id (tool invocation ID)

These are left untouched. Only the session-local identifiers (sessionId, uuid, parentUuid, agentId, cwd) are regenerated.

Restoring After Import

If an import causes problems:

python3 restore_backup.py           # Show snapshot info
python3 restore_backup.py --restore # Restore pre-import state

Restore requires typing "RESTORE" to confirm. The --yes flag bypasses this for automation.

Path Normalization

Claude Code normalizes project paths for storage:

Character	Replacement
/	-
\	-
:	-
.	-
_	-

Unix paths are prefixed with -. Windows paths are not.

Examples:

• /mnt/c/python/my_project becomes -mnt-c-python-my-project
• C:\Users\dev\project becomes C-Users-dev-project

cctrace replicates this normalization to locate and import sessions correctly.

Backwards Compatibility

Existing cctrace users are unaffected:

• The default export behavior is unchanged
• All original output files are still generated
• Portable exports include the legacy files alongside the new structure
• The --in-repo flag is required to use the new portable format

Installation

Requires Python 3.6+ and access to ~/.claude/projects/.

git clone https://github.com/jimmc414/cctrace.git
cd cctrace
./setup.sh

Or manually:

cp export_claude_session.py ~/claude_sessions/
cp import_session.py ~/claude_sessions/
cp restore_backup.py ~/claude_sessions/
mkdir -p ~/.claude/commands
cp .claude/commands/*.md ~/.claude/commands/

Usage

Export

# Classic export to ~/claude_sessions/exports/
python3 export_claude_session.py

# Portable export to .claude-sessions/
python3 export_claude_session.py --in-repo --export-name feature-work

# Export specific session
python3 export_claude_session.py --session-id f33cdb42-0a41-40d4-91eb-c89c109af38a

# Adjust active session detection window (default: 300 seconds)
python3 export_claude_session.py --max-age 600

Import

# Import from local export
python3 import_session.py .claude-sessions/my-session/

# Skip config file import
python3 import_session.py .claude-sessions/my-session/ --skip-config

# Skip auxiliary files (file-history, todos, plans)
python3 import_session.py .claude-sessions/my-session/ --skip-auxiliary

# Non-interactive mode
python3 import_session.py .claude-sessions/my-session/ --non-interactive

Slash Commands

If installed to ~/.claude/commands/:

/export-session [name]    # Portable export
/import-session <path>    # Import
/restore-backup           # Restore from snapshot

Session Detection

When multiple sessions exist for a project, cctrace identifies the correct one by:

1. Finding all .jsonl files in the project's Claude storage directory
2. Filtering to sessions modified within --max-age seconds
3. If running inside Claude Code, correlating the parent PID with session activity
4. Falling back to the most recently modified session

Output Formats

JSONL (raw_messages.jsonl, session/main.jsonl)

The original session format. Each line is a JSON object representing a message or event. Contains all fields exactly as stored by Claude Code.

Markdown (conversation_full.md, RENDERED.md)

Human-readable conversation with:

• Timestamps per message
• User and assistant messages clearly delineated
• Thinking blocks in collapsible <details> sections
• Tool uses with JSON inputs and results

RENDERED.md includes a header with session metadata and is formatted for GitHub rendering.

XML (conversation_full.xml)

Structured format preserving:

• Message hierarchy via UUID/parent-UUID relationships
• Token usage per message
• Tool execution metadata (response codes, duration)
• Thinking block signatures

Manifest (.cctrace-manifest.json)

Required for import. Contains:

• cctrace version
• Original session ID and slug
• Export timestamp
• Original environment context (user, platform, paths)
• Inventory of exported files

File Locations

Path	Purpose
~/.claude/projects/<path>/	Claude Code session storage
~/claude_sessions/exports/	Classic export destination
.claude-sessions/	Portable export destination (in repo)
~/.claude-session-imports/	Import logs and snapshots
~/.claude/commands/	User slash commands

Limitations

• Sessions are one-way portable: you can export and import, but there's no merge or sync
• File-history snapshots reference files by content hash; the original paths are not preserved
• Import generates new UUIDs, so the imported session is a copy, not the original
• Large sessions (500+ messages) produce large RENDERED.md files that may be slow to render on GitHub

Tests

python3 -m pytest tests/ -v

33 tests covering manifest validation, UUID regeneration, path normalization, and restore functionality.

License

MIT