Supported Agents

Traces uses adapter modules in the CLI to read each agent's native storage format and normalize sessions into a common trace structure.

Compatibility matrix

How adapters work

When you run traces share, the CLI:

1. Auto-detects which agent was used (or uses the --agent flag)
2. Locates session data from known storage paths
3. Normalizes data into typed trace messages
4. Uploads the result through the API

Adapter path overrides

Most adapters use platform defaults, but you can override storage paths with environment variables when your runtime layout differs.

Cursor in Docker or remote dev containers

If Cursor is running on the host machine and traces is running inside Docker, bind-mount Cursor's globalStorage directory and set TRACES_CURSOR_GLOBAL_DB to the database path inside the container:

volumes:
  - ~/Library/Application Support/Cursor/User/globalStorage:/cursor-global-storage
environment:
  TRACES_CURSOR_GLOBAL_DB: /cursor-global-storage/state.vscdb

This ensures the Cursor adapter reads from the mounted database location instead of platform defaults.

Agent detection

The CLI detects agents by checking signature files and directories:

• Claude Code -- .claude/
• Cursor -- Cursor storage databases and project markers
• OpenCode -- .opencode/ or XDG data directories
• Codex -- .codex/ or codex.json
• Gemini CLI -- .gemini/ directories
• Pi -- .pi/ directories
• Amp -- .amp/ and XDG data directories
• Copilot -- .copilot/ and Copilot instructions markers
• Cline -- .cline/ and Cline rules files

If multiple agents are detected, the CLI prompts you to choose. Skip detection when needed:

traces share --agent claude-code

Related commands

• traces setup installs share skills/hooks for detected agents
• traces remove skills removes installed sharing setup
• traces doctor validates detection and setup health
• traces share --list --agent <id> helps you select the correct session