agents-cli-aqua

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • Infrastructure Permission Management: The skill utilizes Terraform and gcloud to manage IAM policies and service account permissions, including configurations like TF_VAR_act_as_grant_scope=project. These operations are required for deploying and attaching the AQuA infrastructure.
  • Processing of Production Telemetry: The skill accesses production telemetry from BigQuery and Cloud Logging to retrieve conversation traces and insights for analysis. This data is handled according to the skill's purpose of observability.
  • Indirect Prompt Injection Surface: The skill ingests untrusted production conversation data to provide automated root-cause diagnosis and suggested code fixes.
  • Ingestion points: Production conversation traces and insights are read using agents-cli aqua get-insight in SKILL.md.
  • Boundary markers: Explicit boundary markers are not defined in the instructions, though user review is encouraged.
  • Capability inventory: The skill can trigger agent redeployments via agents-cli deploy and modify infrastructure via agents-cli aqua attach.
  • Sanitization: The instructions recommend that developers treat proposed edits as a lead rather than a direct patch, implying a manual review stage.
  • Shell Command Execution: The skill frequently uses CLI tools including gcloud, bq, terraform, and jq to perform infrastructure management and data querying tasks.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 03:12 AM