handle-startup-state
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent on how to process
initialState, which can be supplied via untrusted external sources like URL parameters (customAppInternalState). - Ingestion points: The
connectToHostApp()function inSKILL.mdserves as the entry point for external data. - Boundary markers: The provided code implementation uses
try/catchblocks for JSON parsing and merges the result with aDEFAULT_STATEobject, providing a fallback mechanism. - Capability inventory: The skill uses
@cognite/app-sdkfor state synchronization. No dangerous operations likeeval()or direct system commands are performed on the ingested data. - Sanitization: Defensive
JSON.parseusage is mandated to ensure that malformed or malicious strings do not crash the application environment.
Audit Metadata