iam-review
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: Static analysis triggers were identified on phrases such as "ignore previous instructions." These are false positives as they are part of explicit defensive instructions telling the agent how to handle potential attacks found in the analyzed data rather than being attacks themselves.
- [DATA_EXFILTRATION]: The skill contains explicit instructions to prevent data exfiltration. It mandates that any credentials, access keys, or secrets found during the assessment must not be included in the output and should be redacted or referenced generically.
- [COMMAND_EXECUTION]: The skill's environment is restricted to read-only operations via the
allowed-toolsconfiguration (Read, Grep, Glob). The instructions emphasize that the skill is for assessment only and must not be used to execute IAM policy changes or any other administrative actions. - [EXTERNAL_DOWNLOADS]: No external dependencies, remote scripts, or package installations are utilized or referenced in the skill definition.
- [PROMPT_INJECTION]: The skill includes a dedicated 'Injection Hardening' section and a 'Prompt Injection Safety Notice' that proactively instruct the agent to treat all processed data as untrusted input, effectively mitigating risk from indirect prompt injection vectors found in IAM metadata.
Audit Metadata