The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a defensive script scripts/search_references.py that utilizes a sanitize_body function. This function proactively filters out common prompt injection patterns (e.g., 'ignore previous', 'you must') from reference files before the content is presented to the agent.
[DATA_EXFILTRATION]: The Python utility implements a path traversal guard in the _validate_refs_dir function. This ensures that file access is strictly confined to the skill's root directory, preventing unauthorized access to the host file system.
[COMMAND_EXECUTION]: All provided scripts use standard Python libraries for file parsing and regex operations. No usage of dangerous functions such as eval(), exec(), or subprocess was detected.
[SAFE]: The documentation and code samples align with the stated purpose of managing Webflow Analyze and Optimize settings. No malicious obfuscation, hidden persistence mechanisms, or unauthorized data access patterns were identified.

webflow-browser-api