incident-response
Warn
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to perform various system operations, including running AWS CLI commands, Kubernetes commands (kubectl), and file system manipulations. It interpolates external input (Alarm ID) into shell variables for alarm lookup. - [EXTERNAL_DOWNLOADS]: References specific pinned versions of MCP servers from the official AWS Labs (
awslabs) repository. This is used to extend the agent's monitoring capabilities with CloudWatch and Prometheus integration. - [DATA_EXFILTRATION]: Performs network requests using
curlto PagerDuty's API to trigger on-call notifications. This is a functional requirement for incident response and uses an environment variable for authentication. - [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: Reads alarm payloads, including descriptions and tags, from CloudWatch and Prometheus alerts (SKILL.md Step 1).
- Boundary markers: No explicit boundary markers or instructions to ignore embedded commands within the alarm data are provided.
- Capability inventory: The skill has access to shell execution, Kubernetes management, and network operations (SKILL.md Step 4 and 5).
- Sanitization: While it performs basic character replacement for filenames, it does not fully sanitize the alarm data used to generate the hypotheses or remediation scripts.
Audit Metadata