Skip to main content

Claude Cowork best practices

Last updated 1 min read

Claude Cowork logs are ingested the same way as Claude Code: from locally saved session transcripts rather than live API interception. The same constraints apply, so latency and temperature are always blank, while prompts, responses, and token counts arrive in full once a session ends.

Same ingestion mechanics as Claude Code

Cowork logs come from locally-saved session transcripts, not live API interception. All the same constraints apply: latency and temperature are never available.

Differences from Claude Code logs

Unlike Claude Code Best Practices, Cowork logs do not populate:

  • Background Agent — always false for Cowork logs; the background-agent detection logic is specific to Claude Code sessions
  • User Turn Count — always blank for Cowork logs; turn-counting is not applied to Cowork transcripts

Caching

Both Cached Input Tokens and Cache Creation Tokens are captured from transcript metadata when present — same as direct Anthropic API logs.

Structured outputs

Not applicable — Cowork logs are session transcripts, not configurable API calls.

Reasoning tokens

Same as Claude Code Best Practices: not separated from output tokens; thinking content appears in Thinking Response but the token count is not broken out.

Latency

Always blank. Session transcripts do not include per-turn timing information.

Related articles