The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines workflows that ingest untrusted system data, creating an attack surface for indirect prompt injection.
Ingestion points: The runbooks instruct the agent to read output from kubectl logs, kubectl describe, and journalctl, which may contain attacker-controlled content.
Boundary markers: The provided templates do not utilize specific delimiters or instructions to ignore embedded prompts in external data.
Capability inventory: The instructions involve the use of powerful administrative tools including kubectl, psql, and ssh for diagnostics and mitigation.
Sanitization: There are no explicit sanitization or validation steps for data retrieved from logs or system events.
[SAFE]: The skill consists of legitimate operational documentation and standard troubleshooting procedures. No evidence of credential exfiltration, obfuscation, or unauthorized remote code execution was found.

sre-runbooks