Interface: ToolCallFacts

Defined in: packages/security/src/policy/tool-argument-policy.ts:36

Facts about a tool call the policy engine decides over.

Properties

Property	Modifier	Type	Description	Defined in
`args?`	`readonly`	`unknown`	The validated arguments (post-schema, post-repair).	packages/security/src/policy/tool-argument-policy.ts:51
`sensitive?`	`readonly`	`boolean`	`true` when the tool reads/handles sensitive data (e.g. `sensitivity: 'secret'`).	packages/security/src/policy/tool-argument-policy.ts:40
`sideEffectClass`	`readonly`	`PolicySideEffectClass`	-	packages/security/src/policy/tool-argument-policy.ts:38
`toolName`	`readonly`	`string`	-	packages/security/src/policy/tool-argument-policy.ts:37
`untrustedSource?`	`readonly`	`boolean`	`true` when the tool is an untrusted-content SOURCE - its trust class is one the taint engine treats as injection-bearing (mcp-derived / web-search / skill-untrusted; see `isUntrustedTrustClass` in `@graphorin/security/dataflow`). Powers the Rule-of-Two `untrustedInput` leg. The engine stays pure: the caller derives this from the tool's metadata.	packages/security/src/policy/tool-argument-policy.ts:49