The Agent Skills Directory

Safe Data Handling (Indirect Prompt Injection Mitigation): The skill is designed to ingest external data including CloudTrail logs, IAM policy documents, and user-supplied error messages. It proactively mitigates potential indirect prompt injection risks by providing the agent with explicit instructions to treat all collected content as untrusted data and to ignore any instructions found within policies, tags, or log fields.
Read-Only Operation Enforcements: The skill implements a strict read-only posture by maintaining an allowlist of permitted AWS API operations (such as Get*, List*, and Describe*) and explicitly forbidding any mutating actions like Put*, Attach*, or Delete*. This behavior is further reinforced by the agent's platform-level security guardrails.
Metadata and Identifier Exposure: To fulfill its diagnostic purpose, the skill processes and reports metadata such as Resource ARNs, Account IDs, and Role names. The skill includes appropriate warnings regarding the sensitivity of this metadata, treating it with the same care as IAM configuration data.
Trusted External References: All external resources and documentation links point to official AWS domains (e.g., docs.aws.amazon.com, repost.aws) or trusted GitHub organizations (github.com/aws). No unverifiable third-party dependencies or remote code execution patterns were detected.

aiml-access-diagnostics