google-agents-cli-observability
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- Indirect Prompt Injection Surface: The skill describes implementing a feedback mechanism in
references/feedback-mechanism.mdthat ingests arbitrary user text. - Ingestion Point: Untrusted user input enters the system through the
/feedbackFastAPI endpoint into theFeedbackmodel'stextfield. - Capability Inventory: Ingested data is stored in BigQuery and Cloud Logging, where it is intended to be used for conversational analytics and LLM-as-judge evaluations.
- Boundary Markers: There are no specific instructions or delimiters provided to prevent the agent from interpreting instructions embedded within the feedback text during subsequent analysis.
- Sanitization: While the skill advises on PII redaction, it does not provide methods for sanitizing the input against adversarial prompt instructions.
- Command Execution: The skill involves the use of
agents-cli,bq, andgsutilfor infrastructure management and data verification. These are part of the skill's primary functionality for managing Google Cloud resources and do not represent unexpected behavior.
Audit Metadata