The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the monitor() function which allows agents to execute shell commands (e.g., make test, pytest -q) in a separate process group. Merging stderr into stdout is a standard feature of this utility.
[DATA_EXPOSURE]: The documentation includes the tail() function which enables reading and streaming local files (e.g., app.log) directly into agent input channels.
[INDIRECT_PROMPT_INJECTION]: The reactive input system creates an inherent surface for indirect prompt injection by piping external data into the agent's context.
Ingestion points: External data flows into the agent via monitor() (process output), tail() (file content), cron() (timers), and QueueManager queues or events.
Boundary markers: The skill uses QueueOutput events to render metadata previews in the prompt, which provides some separation, though the full data remains accessible to the underlying logic.
Capability inventory: The associated capabilities include shell command execution via monitor(), file reading via tail(), and asynchronous task execution via spawn().
Sanitization: There is no mention of sanitization, filtering, or validation for the content ingested through these channels.

nooa-channels