Capture with Claude Code

Run Claude Code through paper so the session is captured. Prerequisites: paperctl installed and authenticated.

Start

paperctl start claude

That’s the whole path. paperctl start claude sets ANTHROPIC_BASE_URL for the launched process and starts Claude Code through the local proxy. Use Claude Code normally from there — your existing Claude Code sign-in works as-is.

Anything after -- is passed straight through to Claude Code, so paper’s flags never collide with the agent’s:

paperctl start claude -- --dangerously-skip-permissions

To make plain claude route through paper automatically, set up shell routing — then you never have to remember the wrapper.

Verify the capture

After a session, confirm it was recorded:

paperctl sessions list

The session appears at the top with its turn count, cost, and status, and shows up in paper console moments later. Subagent runs — Task agents, however many run in parallel — appear inside the same session, each nested under the tool call that spawned it, and the session’s usage and cost totals include them.

If something’s off

• No session in the list: check paperctl status (daemon running? auth healthy?) and paperctl logs.
• Session appears but looks incomplete: the final transcript push happens when the agent exits — give it a minute.
• More cases: CLI troubleshooting.