Graphorin API reference v0.12.1
Graphorin API reference / @graphorin/tools / / ExecuteBatchOptions
Interface: ExecuteBatchOptions
Defined in: packages/tools/src/executor/types.ts:293
Per-batch options accepted by executor.executeBatch(...).
Properties
| Property | Modifier | Type | Description | Defined in |
|---|---|---|---|---|
calls | readonly | readonly ToolCall[] | - | packages/tools/src/executor/types.ts:294 |
capability? | readonly | "read-only" | Run-level capability restriction (D2 - single-writer constraint). 'read-only' deterministically blocks every side-effecting / external-stateful tool with a capability_blocked outcome, no matter what the model asked for - the enforcement half of the agent-side advertise filter. Absent ⇒ all capabilities (legacy). | packages/tools/src/executor/types.ts:323 |
disableRepair? | readonly | boolean | Disable the single-round repair hook for this batch (tools-02). Used for PRE-APPROVED calls replayed on a durable-HITL resume: a human granted exactly these args, so a repair rewrite must fail as invalid_input rather than execute a payload nobody saw. | packages/tools/src/executor/types.ts:305 |
preApproved? | readonly | boolean | E1: this batch replays calls a human ALREADY granted through the agent's durable-HITL pre-screen. ask/defer verdicts from the permission hook / argument policy are treated as satisfied (the grant IS their resolution) instead of failing closed; deny still blocks, and a hook rewrite of the granted args fails the call (tools-02). Set together with disableRepair by the agent's resume dispatch. | packages/tools/src/executor/types.ts:315 |
runContext | readonly | RunContext | - | packages/tools/src/executor/types.ts:295 |
stepNumber | readonly | number | - | packages/tools/src/executor/types.ts:296 |
trustLevel? | readonly | SandboxTrustLevel | Trust level for the per-tool sandbox resolution. Default 'user-defined'. | packages/tools/src/executor/types.ts:298 |