The Agent Skills Directory

Indirect Prompt Injection (HIGH): This skill exhibits a dangerous combination of untrusted data ingestion and high-privilege capabilities.
Ingestion points: The skill reads external content through grep on /var/log/app.log and kubectl logs. Logs frequently contain data from user-supplied request headers, error messages, and stack traces.
Boundary markers: There are no delimiters or 'ignore' instructions specified to prevent the agent from obeying instructions embedded within the logs.
Capability inventory: The skill allows the agent to execute kubectl rollout undo, kubectl scale, and curl -X POST to administrative endpoints. This allows for significant modification of production infrastructure.
Sanitization: No sanitization or filtering of log content is implemented.
Command Execution (HIGH): The skill provides the agent with direct access to production-grade tools (kubectl, psql, curl) with write access. While necessary for the stated purpose of 'incident response', the lack of human-in-the-loop checkpoints for infrastructure modification is a significant security risk.

incident