The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and analyze untrusted telemetry data, including process command lines and raw log contents. This presents a potential surface for indirect prompt injection if malicious instructions are embedded within the logs. The skill proactively addresses this by including a "Prompt Injection Defense Directive" that instructs the agent to treat all retrieved data strictly as untrusted and never as instructions.
Dynamic Command and Query Generation: The skill generates structured UDM queries and shell commands (e.g., gcloud) based on hunt hypotheses and external indicators. While these are necessary for the skill's functionality, users should ensure the agent's generated queries are bounded (as instructed in the skill) and that shell commands are reviewed before execution.
Usage Attribution Tracking: The skill includes instructions to include attribution tags in environment variables and HTTP headers when executing commands. This is used for vendor metrics and does not involve the exfiltration of user-sensitive data, but it is a behavioral pattern to be aware of during execution.

secops-hunt