The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains defensive instructions that explicitly warn the agent to ignore directives embedded in external data (such as fetched pages or screenshots) and to treat such content as data rather than instructions. A static detector hit was identified as a false positive on these safety guidelines.
[CREDENTIALS_SAFE]: Code snippets for advertising tracking pixels use standard placeholders (e.g., 'YOUR_PIXEL_ID', 'AW-XXXXXXXXX'). No hardcoded credentials or sensitive API keys are present.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges the risk of ingesting untrusted external data. It defines clear ingestion points (e.g., marketing context files, screenshots, and account exports) and provides mandatory safety instructions to treat this data as non-authoritative. This correctly addresses the indirect injection attack surface.
[REMOTE_CODE_EXECUTION]: There are no instances of remote script execution, package installation, or dynamic code generation. The skill primarily consists of instructional documentation.
[DATA_EXFILTRATION]: No patterns of sensitive file access or network exfiltration to unauthorized domains were found. The skill's purpose is aligned with its functionality.

ads