warden-iam
Pass
Audited by Gen Agent Trust Hub on Jul 1, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill reads infrastructure-as-code files and cloud configurations (e.g.,
*.tf,Pulumi.*, AWS/GCP/Azure configs) from the local environment to map access requirements. These files represent untrusted data that could contain embedded instructions intended to manipulate the agent's behavior. - Ingestion points: Processes configuration and IaC files found in the project environment to identify existing IAM and services (SKILL.md, Step 0).
- Boundary markers: The prompt lacks instructions to wrap ingested content in delimiters or to ignore instructions found within those files.
- Capability inventory: The agent has access to powerful tools including
Bash,Write,Edit,WebFetch, andWebSearch(SKILL.md, frontmatter). - Sanitization: No validation or sanitization of the input data is performed before processing.
- Output Suppression: The skill instructs the agent to suppress detailed analysis from the CLI and only provide it via an external report (
/atlas-report), which reduces the user's ability to monitor the agent's reasoning and detect potential instruction overrides in real-time.
Audit Metadata