The Agent Skills Directory

[SAFE]: The skill implements proactive defense against prompt injection by instructing the agent to ignore and report any directives (such as "ignore previous instructions") found within analyzed IAM policies or metadata.
[SAFE]: The skill enforces strict data handling boundaries, explicitly forbidding the execution of commands found in policies and the exfiltration of any discovered credentials or secrets.
[SAFE]: Access is restricted to read-only tools (Read, Grep, Glob), and the skill includes a clear directive that it is for assessment only and must not perform configuration changes.
[SAFE]: No external dependencies, remote code downloads, obfuscation, or persistence mechanisms were detected in the skill's instructions.

iam-review