Skip to content

Graphorin API reference v0.12.1


Graphorin API reference / @graphorin/security / / DataFlowEvaluation

Interface: DataFlowEvaluation

Defined in: packages/security/src/dataflow/types.ts:257

The signal a DataFlowPolicy evaluates for one candidate sink call. Populated by the enforcement point from the resolved tool's metadata plus the run's TaintLedger.

Stable

Properties

PropertyModifierTypeDescriptionDefined in
carriesUntrustedVerbatimreadonlybooleantrue when the sink's arguments carry untrusted content verbatim.packages/security/src/dataflow/types.ts:273
sensitiveSeenreadonlybooleantrue when secret-tier content has entered the run.packages/security/src/dataflow/types.ts:277
sideEffectClassreadonlySideEffectClassThe sink's resolved side-effect class.packages/security/src/dataflow/types.ts:261
sinkKind?readonly"tool" | "assistant-output"B4 (item 14): what KIND of sink this evaluation describes. 'tool' (default when absent) - a tool call gated by its side-effect class. 'assistant-output' - the run's outgoing assistant text, a sink by definition regardless of side-effect class (the reply surface exfiltrates to whoever reads it). Declassify matching stays on toolName (e.g. the stable id 'assistant-output' in declassifySinks).packages/security/src/dataflow/types.ts:271
sourceKindsreadonlyreadonly string[]Untrusted source kinds relevant to this flow (matched + observed).packages/security/src/dataflow/types.ts:279
toolNamereadonlystringName of the sink tool about to run (the stable sink id).packages/security/src/dataflow/types.ts:259
untrustedSeenreadonlybooleantrue when untrusted content has entered the run.packages/security/src/dataflow/types.ts:275