incident-response-incident-response
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data through the $ARGUMENTS variable across multiple processing stages without boundary markers or sanitization.
- Ingestion points: The $ARGUMENTS placeholder is interpolated into prompts for every subagent task defined in SKILL.md.
- Boundary markers: There are no delimiters or 'ignore embedded instructions' warnings to separate external incident data from core instructions.
- Capability inventory: The workflow orchestrates subagents with capabilities for system debugging, security assessment, fix implementation, and production deployment.
- Sanitization: No escaping, validation, or filtering mechanisms are defined for the content passed into subagent prompts.
- [COMMAND_EXECUTION]: The skill instructs the agent to perform high-privilege operational actions including resource scaling, circuit breaker activation, and production deployments via specialized subagents. While these are intended for incident resolution, they represent a significant capability surface.
Audit Metadata