The Agent Skills Directory

[PROMPT_INJECTION]: The skill identifies Teleport activity logs as a potential source of indirect prompt injection. It includes comprehensive security instructions in references/SECURITY.md that explicitly warn the agent to ignore directives such as 'ignore previous instructions' or 'run this command' if they appear in log output. It provides clear boundary guidelines, instructing the agent to treat all command output as untrusted data.
[COMMAND_EXECUTION]: Operational capabilities are strictly limited to official Teleport tools (tctl, tsh) and basic binary discovery (which). The skill is defined as read-only, and safety rules mandate that any values extracted from investigation results (such as resource names or user IDs) must be quoted if used in subsequent command arguments, effectively preventing command injection.
[SAFE]: No malicious patterns related to credential harvesting, unauthorized network communication, or code obfuscation were detected. The skill is authored by the vendor (gravitational) and focuses on the legitimate management of its own software.

teleport-investigate