The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where untrusted external data could influence the agent's actions.
Ingestion points: Multiple instructions in SKILL.md direct the agent to load content from local files such as dynamo-policy.json, trust.json, and waf-rules.json as arguments for AWS commands.
Boundary markers: No delimiters or instructions to ignore potential commands within these files are provided.
Capability inventory: The skill uses the shell tool to execute aws CLI commands, which possess the capability to modify cloud resources and policies.
Sanitization: No validation or sanitization is performed on the content of the external files before they are processed by the shell environment.
[COMMAND_EXECUTION]: The skill utilizes the shell tool to perform various AWS administrative and security tasks. These include creating IAM roles, managing KMS keys, and configuring WAF rules, which are all consistent with the skill's stated purpose as a security reference.
[DATA_EXFILTRATION]: The skill includes instructions to generate and decode IAM credential reports (aws iam get-credential-report). While this action exposes sensitive account security metadata to the agent's execution context, it is a standard auditing procedure and no instructions for unauthorized external transmission were found.

aws-security