The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a recurring prompt mechanism that accepts natural language instructions, creating a surface where malicious data could influence future agent behavior.
Ingestion points: The instruction parameter in the create and update methods within src/rlm_heartbeat/__init__.py accepts arbitrary strings intended for future agent turns.
Boundary markers: The implementation does not utilize explicit delimiters or security markers to isolate these scheduled instructions from the rest of the agent's context.
Capability inventory: The skill is limited to heartbeat lifecycle management (list, create, update, delete) via the platform-provided host_request bridge. It lacks any functionality for subprocess execution, file system writes, or network operations.
Sanitization: The code performs standard Python type checking for its parameters but does not implement any content-based filtering or sanitization of the provided instructions.

rlm-heartbeat