Skip to main content

​
Recap

cmd/crewship/cmd_recap.go produces a structured summary of an existing chat session by dispatching the transcript through the default agent. Useful when picking up an older thread, sharing a session with a teammate, or auditing what an agent actually did across many turns. Pulls up to 500 messages (same cap as crewship chat), compresses tool-call noise to names only, truncates each turn at 800 characters to keep prompt tokens bounded, then re-uses crewship ask’s streaming pipeline so output renders incrementally.

​
crewship recap <chat-id>

FlagTypeDefaultEffect
--agent <slug>stringdefault-agentAgent slug used to generate the summary. Falls through to default_agent config when unset.
--bullets <n>int8Cap on the “Key decisions” section.

​
Examples

crewship recap c_abc123
crewship recap c_abc123 --agent viktor
crewship recap c_abc123 --bullets 5    # shorter

​
Output shape

The summarising agent is instructed to produce a markdown block with these sections, in order, with no preamble:
  1. Final outcome — one sentence
  2. Key decisions — up to --bullets items
  3. Open threads — anything unresolved, or (none)
  4. What to ask next — one suggested follow-up prompt

​
Common errors

  • chat <id> has no messages — the chat is empty or doesn’t exist.
  • internal: ask command has no RunE — wiring error; should not appear in shipped builds.

​
See also