The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected in the skill's logic or instructions.
[COMMAND_EXECUTION]: The skill directs the agent to use standard cloud provider CLI tools (AWS, GCP, Azure) for auditing tasks. It includes a 'Boundaries' section that explicitly limits the agent to read-only operations and forbids any actions related to bypassing MFA or social engineering.
[PROMPT_INJECTION]: The skill contains explicit instructions for the agent to refuse requests aimed at bypassing security controls, such as credential recovery via social engineering or designing backdoor access.
[EXTERNAL_DOWNLOADS]: The skill references established technology services and well-known authorization tools (e.g., Cerbos, OpenFGA, Oso, Okta, Auth0). These represent standard industry solutions for identity management.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external cloud configuration data, it implements safeguards through structured reporting requirements and read-only constraints.
Ingestion points: IAM policies, audit logs, and configuration details retrieved from cloud providers (SKILL.md).
Boundary markers: The instructions mandate the use of a specific executive summary and findings report schema.
Capability inventory: Bash, Read, Write, Grep, Glob, and WebSearch tools are available to the agent.
Sanitization: The skill relies on authenticated CLI commands for data retrieval and focuses on analytical reporting rather than the execution of untrusted payloads.

iam-audit