The Agent Skills Directory

[Command Execution]: The skill utilizes several AWS CLI commands (such as aws cleanrooms, aws iam, aws s3api, aws kms, and aws logs) to retrieve configuration data.
Evidence: Commands like aws iam get-role-policy and aws s3api get-bucket-policy are used in references/permission-debugging.md and references/custom-model-logging-debugging.md to inspect security policies.
Context: These operations are focused on retrieving read-only configuration metadata to identify root causes for access failures. This is standard procedure for cloud infrastructure troubleshooting.
[Indirect Prompt Injection]: The skill accepts user-supplied error messages as input to guide the classification of permission failures.
Ingestion points: The error_message parameter in references/permission-debugging.md is provided by the user.
Capability inventory: The skill possesses extensive read capabilities for AWS resources using the CLI.
Sanitization: The skill uses the input for classification logic rather than direct interpolation into executable command strings, which limits the potential impact of adversarial content.
Context: While processing untrusted data is a known surface for indirect prompt injection, the skill's logic uses this data to select appropriate diagnostic paths, maintaining a human-in-the-loop approach for generating a final diagnosis.

aws-cleanrooms