Agent onboarding
What you'll do
Run craik onboard before starting project work. The command emits a
runner-parseable craik.agent_onboarding JSON report covering the
project model, active policy, docs boundaries, recent handoffs,
contradictions, stale-risk warnings, validation commands, Stigmem
status, known traps, and allowed next actions.
Onboarding is read-only.
The command prints structured JSON. It does not modify state, make live Stigmem calls, or surface secrets.
Run it
craik onboard --project Example
The output is structured JSON safe for runners to parse directly.
Policy profile
craik onboard --project Example --policy-profile strict
stricttrusted-localcraik onboard --project Example --policy-profile trusted-local --trusted-local-fail-open.Documentation boundaries
Immutable paths are evidence.
The docs_boundaries section separates mutable docs from immutable
paths. Immutable docs (such as ADRs) are evidence — do not edit them
unless a separate policy approval explicitly allows it.
Continuity checks
Onboarding surfaces six continuity fields. Agents must review them before creating a plan.
Recent handoffs
For the project.
Unresolved contradictions
Stale-risk warnings
Validation commands
Configured for the project.
Known traps
Allowed next actions
Missing handoffs, missing docs paths, dirty repository state, unresolved contradictions, or missing Stigmem environment configuration are reported as stale-risk warnings.
Stigmem status
Onboarding does not print secrets or make a live Stigmem request. For
Stigmem projects, it reports whether CRAIK_STIGMEM_URL and
CRAIK_STIGMEM_API_KEY are configured. Use
craik connect stigmem when a live backend compatibility check is
needed.