Human delegation and scope changes

2 min readReferenceUpdated 2026-05-19

What you'll find here

The two contracts that bound where autonomous agents must stop — craik.human_delegation_point for delegated decisions and craik.scope_change_request / _result for changes to accepted task scope.

Open delegation blocks autonomy.

Open delegation points and pending scope-change requests block autonomous continuation. Resolutions must be durable and auditable.

Delegation kinds

Kind

Use for

Trigger

approval

authorization

Human authorization is required by policy or task boundary.

clarification

ambiguity

The request is ambiguous enough that guessing would change behavior or scope.

escalation

unresolved risk

An unresolved risk, contradiction, or policy issue needs a human decision.

ownership_transfer

handover

The agent should transfer responsibility to a human or another owner before further work.

Resolved delegation points must include resolution text.

Scope changes

Contract

Records

Lifecycle

craik.scope_change_request

request

Current scope · proposed scope · reason · intent lock · optional policy envelope · contradictions · handoffs.

craik.scope_change_result

decision

Records the human decision. Rejected changes keep the existing scope. Accepted changes must link to an updated intent lock so the new boundary is durable and auditable.

Handoffs

Handoffs surface open_human_delegation_ids, scope_change_request_ids, and scope_change_result_ids. A blocked handoff must list open delegation points and explain what human input is required before work resumes.

What's next

ReferenceDelegation queue viewThe operator surface that displays open delegations.ReferenceSchema referenceThe full delegation and scope-change schemas.GuideScope controlHow intent locks and stop-conditions compose with delegation.