inngest

No direct evidence of malware or covert data theft appears in the supplied fragment. The implementation is an agent orchestration pattern, but it presents a significant conditional risk because model-controlled tool names and inputs are passed directly to executeTool without visible allowlisting, validation, authorization, or human approval. The omitted tools and dispatcher must be reviewed before deployment, especially for destructive or external side-effecting operations. Sending conversation and tool data to Anthropic is an explicit, expected network flow that requires appropriate data-handling controls.