Skip to main content

​
crewship persona

Manage PERSONA.md — the ~1.5 KB identity surface the orchestrator injects into every agent’s system prompt at session start. The CLI mirrors the agent persona API and the crew persona API. Two layers:
  • Agent layer — per-agent override. Wins outright when non-empty.
  • Crew default layer — workspace-shared baseline that every agent in the crew falls back to.
All commands operate on the active workspace (crewship workspace use) and resolve targets by slug, not ID, to match the rest of the CLI surface.

​
Subcommands

​
crewship persona view

Prints the resolved persona — agent-layer file if present, otherwise crew default, otherwise the per-role synthesised stub — along with a header line showing the source layer and byte count, then the content.
The header’s source is the resolved layer (agent or crew). When neither agent nor crew has a file on disk and the response is the You are the {role}... synthesised text (from_default: true on the wire), the source reads synthesized default — useful when reviewing a freshly-onboarded crew.

​
crewship persona edit

Pulls the current agent-layer content, opens $EDITOR (falling back to vi), and writes the saved buffer back via PUT /api/v1/agents/{id}/persona. An empty buffer aborts the edit without writing.
The server caps persona content at 1500 bytes.

​
crewship persona reset

Removes the agent-layer file. The next view falls through to the crew default (or the synthesised stub when the crew has none).

​
crewship persona history

Lists memory_versions rows for the agent’s PERSONA.md path, newest first (most recent 20 entries). Use the printed sha256 with crewship memory show to read an older version’s content without restoring it.

​
crewship persona suggest-from-inbox

This is a Phase-2 stub. When an agent’s POST /persona/suggest lands as an inbox proposal (because the crew’s autonomy is strict / guided / trusted), there is not yet a dedicated endpoint to fetch an arbitrary audit row by id, so the command does not auto-apply. It prints a hint to view the proposal in the inbox UI and apply it manually via crewship persona edit <agent>.

​
crewship persona crew

The verb comes after the crew slug (crewship persona crew <crew-slug> <verb>), and both arguments are required — the command takes a minimum of two positional args and errors with a usage hint if the verb is omitted. Same verbs against the crew-default layer at /api/v1/crews/{crewId}/persona. There is no suggest equivalent — agents propose only into their own layer, and the operator authors crew defaults directly.

​
See also

  • Agent Persona API — REST shape per endpoint, plus the policy gating for suggest.
  • Crew Persona API — the default layer this CLI’s crew sub-tree edits.
  • Autonomy & self-learning — how crew autonomy_level decides whether an agent’s suggest auto-applies or queues for operator approval.