investigating-with-observability
Pass
Audited by Gen Agent Trust Hub on Mar 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the Bash tool with curl to perform network requests to VictoriaMetrics, VictoriaLogs, and VictoriaTraces APIs using URLs and authentication headers provided in environment variables.
- [PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks (Category 8) because it ingests and processes data from external sources such as application logs and traces. Ingestion points: Data enters the context through curl responses from observability backends. Boundary markers: The skill uses subagent partitioning and explicit instructions to 'report raw findings only' to limit interpretation scope. Capability inventory: The skill possesses Bash(curl:*) capabilities which could be targeted by instructions hidden in logs or trace tags. Sanitization: There is no explicit sanitization of log content before it is processed by the agent.
Audit Metadata