The Agent Skills Directory

[PROMPT_INJECTION]: The library implements an IPC communication layer that serves as a vulnerability surface for indirect prompt injection through data received from external processes.\n
Ingestion points: In src/index.ts, the MessageIpc class initializes a listener on the IPC channel using this.channel.on('message', ...).\n
Boundary markers: The implementation lacks boundary markers or instructions to the model to ignore potential instructions embedded within the IPC data payloads.\n
Capability inventory: The library facilitates bi-directional data transfer via the post method in src/index.ts and allows for processing logic in subclass-defined exec methods.\n
Sanitization: No sanitization or validation logic is applied to message payloads before they are passed to the execution handler.

message-ipc