The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes application log files that may contain untrusted data originating from external service responses or user inputs, creating a surface for indirect prompt injection.\n
Ingestion points: Multiple commands in SKILL.md read from /var/log/app/gamma-*.log using grep and tail.\n
Boundary markers: No delimiters or instructions to ignore instructions are used when presenting log content to the agent.\n
Capability inventory: The skill utilizes Bash (specifically curl), Write, and Edit tools, and interacts with internal administrative endpoints on localhost and prometheus.\n
Sanitization: There is no evidence of sanitization, filtering, or validation of the log data before processing.\n- [DATA_EXFILTRATION]: The skill makes network requests to internal services and external domains for monitoring and administrative purposes.\n
Evidence: curl commands are used to query a Prometheus instance at http://prometheus:9090, check health at https://your-app.com/health/gamma, and send POST requests to administrative endpoints on http://localhost:8080 for throttling and circuit breaking.

gamma-incident-runbook