Goal
Define a generic Agents API contract for principal, ownership, and client context that downstream agent runtimes can use for permission-aware tools, citations, and audit trails.
This should support private-site knowledge products without encoding Intelligence or A8C-specific semantics.
Required work
- Clarify session owner, authenticated user, agent owner, workspace, audience, and runtime principal fields.
- Define how client context may explicitly map to tool arguments without broad ambient key matching.
- Provide metadata shape for citations/tool calls that records safe principal/source context.
- Support frontend chat/session surfaces that need to show permission-aware results and diagnostics.
- Document redaction and sensitive-context handling.
Acceptance criteria
- Data Machine and Intelligence can propagate permission context through tool execution and search/read provenance.
- Frontend clients can reason about permission-denied, partial, and source-restricted results.
- The contract remains generic enough for non-Intelligence agents.
Related
Automattic/agents-api#273 for structured pending-action confirmations.
Automattic/intelligence#815 for permission-aware search/read and wiki provenance.
Goal
Define a generic Agents API contract for principal, ownership, and client context that downstream agent runtimes can use for permission-aware tools, citations, and audit trails.
This should support private-site knowledge products without encoding Intelligence or A8C-specific semantics.
Required work
Acceptance criteria
Related
Automattic/agents-api#273for structured pending-action confirmations.Automattic/intelligence#815for permission-aware search/read and wiki provenance.