Skip to main content

​
crewship export

Two distinct things live under export:
  • crewship export crew <slug> / crewship export workspace — render the live workspace as a YAML manifest you can commit to git, share with a teammate, or re-apply on another machine.
  • crewship export <run-id> — bundle a single run’s prompt, response, journal entries and metadata into a folder for post-mortems.
crewship export <subcommand|run-id> [flags]
All three forms require an authenticated session and a workspace context (inherited from the current CLI profile, same as crewship apply).

​
crewship export crew

Pull a crew’s current state and render it as a kind: Crew manifest. Round-trip partner of crewship apply: piping the output back through apply on a fresh workspace recreates the same shape. The output is YAML with a yaml-language-server $schema directive on the first line, so editors with the YAML language server (VS Code, JetBrains, Neovim with coc-yaml) give autocomplete and validation while you edit. Pipe through yq -o json if you need JSON. 
crewship export crew <slug> [flags]

​
Flags

FlagDefaultPurpose
-o, --output <path>stdoutWrite to a file instead of stdout.
--no-credentialsfalseStrip credential slots from output (consumers must declare their own).
--no-skill-bodiesfalseSkip skill bodies; emit slug-only references for a structure-only overview.

​
Examples

# Snapshot to stdout
crewship export crew code-review

# Snapshot to a file under version control
crewship export crew code-review -o ./manifests/code-review.crew.yaml

# Structure-only overview (no inline skill bodies)
crewship export crew code-review --no-skill-bodies > overview.yaml

​
What’s included

  • Crew metadata (name, slug, icon, color, description)
  • Devcontainer fields (image, memory, CPUs, mise config)
  • Sidecar services: (Redis, Postgres, etc.) — full shape, including healthchecks and volume names
  • Every agent — slug, prompt, role, LLM provider/model, tool profile, memory flag
  • Each agent’s skills: and env_refs: bindings
  • Skill bodies (inline by default) for every skill bound to an agent
  • Credential slots (without values — these never travel in the export)
  • Crew-scoped MCP servers (crew_integrations)

​
What’s not included

  • Credential values (intentional — manifests are safe to commit)
  • Computed fields (IDs, timestamps, cached image hashes)
  • Workspace members (they’re per-user identity, not workspace-shareable)
  • Routines / schedules / inbox / eval scenarios (out of scope for v1 manifests)

​
crewship export workspace

Render every crew in the active workspace as a single kind: Workspace bundle. Workspace-level deduplication is applied: skills and credentials used by any agent in any crew are lifted to the workspace scope so consumers see one declaration each. 
crewship export workspace [flags]

​
Flags

FlagDefaultPurpose
-o, --output <path>stdoutWrite to a file instead of stdout.
--no-credentialsfalseStrip credential slots.
--no-skill-bodiesfalseSkip skill bodies; emit slug-only refs.

​
Examples

# Full backup
crewship export workspace -o ./acme.workspace.yaml

# Re-apply on another instance
crewship export workspace > /tmp/acme.yaml
ssh other-host crewship apply --file - --from-env < /tmp/acme.yaml

​
Dedup behaviour

If three crews each declare house-style, the workspace export emits one skill at workspace scope and per-crew skills: [house-style] references. Same for credentials — ANTHROPIC_API_KEY becomes a single workspace-scope slot regardless of how many agents reference it.

​
crewship export <run-id>

Bundle a single run’s chat, journal, and metadata into a folder. Used for post-mortems, handoffs, or piping the conversation into a different LLM. 
crewship export <run-id> [flags]

​
Flags

FlagDefaultPurpose
--out <dir>./run-<run-id>Output directory.
--no-journalfalseSkip the journal pass.

​
Output layout

<out>/
  run.json       — runMetadata + window
  prompt.md      — first user message recovered from chat
  response.md    — concatenated assistant text
  messages.json  — full chat message list (raw)
  journal.json   — journal entries (oldest-first)
  timeline.txt   — human-readable timeline

​
Examples

# Default directory: ./run-r_abc/
crewship export r_abc

# Custom out path, skip journal
crewship export r_abc --out /tmp/post-mortem --no-journal

​
Round-trip with apply

The manifest exports (crew / workspace) are designed for round-trip: 
crewship export crew code-review > backup.yaml
# ... later, on another machine
crewship apply --file backup.yaml --from-env --yes
A few intentional gaps make this not byte-exact:
  • The exporter strips computed fields (IDs, timestamps).
  • Credential values never travel — pass them via --from-env / --secrets-file on re-apply.
  • URL-fetched skills are re-exported as inline content (the source URL isn’t preserved in the DB row).
For a --dry-run check that confirms the round-trip is clean: 
crewship export crew code-review > round-trip.yaml
crewship apply --file round-trip.yaml --dry-run
The plan should show every resource as = (unchanged).

​
See also