The Agent Skills Directory

[SAFE]: No malicious patterns or unauthorized behaviors were detected. The skill's behavior is consistent with its stated purpose of monitoring external signals.
[PROMPT_INJECTION]: The skill processes data from untrusted external sources (e.g., CI/CD status, endpoint health), which creates an indirect injection surface inherent to its function.
Ingestion points: Output from Bash commands like gh or curl used to query external services.
Boundary markers: The skill does not explicitly define delimiters to separate external data from agent instructions, relying instead on high-level classification logic.
Capability inventory: Uses Bash for state checks and the Agent tool to trigger subsequent actions.
Sanitization: No explicit sanitization or validation of the external signal data is described, as the agent is instructed to read the raw state directly.

watch-and-act