Interface: PermissionRequest

Defined in: src/adapters/types.ts:611

Properties

actor

readonly actor: string

Defined in: src/adapters/types.ts:618

capability

readonly capability: PermissionCapability

Defined in: src/adapters/types.ts:617

What kind of operation is being asked about. See PermissionCapability for which values the framework actually sends and when.

context?

readonly optional context?: Readonly<Record<string, unknown>>

Defined in: src/adapters/types.ts:627

history?

readonly optional history?: readonly LLMMessage[]

Defined in: src/adapters/types.ts:640

v2.12 — Full conversation history at check time. Lets policies inspect prior assistant content / tool results without maintaining parallel state via event subscription.

identity?

readonly optional identity?: object

Defined in: src/adapters/types.ts:650

v2.12 — Caller identity from agent.run({ identity }). Permission predicates can role-check on identity.principal / identity.tenant.

principal?

readonly optional principal?: string

tenant?

readonly optional tenant?: string

iteration?

readonly optional iteration?: number

Defined in: src/adapters/types.ts:645

v2.12 — Current ReAct iteration (1-based). Lets policies fire different rules per iteration without external counters.

sequence?

readonly optional sequence?: readonly ToolCallEntry[]

Defined in: src/adapters/types.ts:634

v2.12 — Sequence of tool calls already dispatched this run, in call order. EMPTY for non-tool_call capabilities. Sequence-aware policies (forbidden chains, idempotency limits) read this to make decisions that single-call governance cannot.

signal?

readonly optional signal?: AbortSignal

Defined in: src/adapters/types.ts:660

v2.12 — Optional abort signal propagated from agent.run({ env: { signal } }). Async checkers (Redis lookups, hub-backed allowlists) MUST honor this — when the agent run is cancelled, in-flight checks should abort.

target?

readonly optional target?: string

Defined in: src/adapters/types.ts:626

What is being asked about, in the vocabulary of the capability:

'tool_call' and every ToolCapability — the TOOL NAME.
'skill_read' — skill:<id> (9.11.0). Prefixed so a skill and a tool of the same name are two different subjects to a policy that lists ids.

PermissionRequest