The Agent Skills Directory

[SAFE]: The skill interacts exclusively with the Webflow ecosystem using official vendor tools such as element_tool and data_sites_tool. All external resources referenced are official documentation from W3C or Webflow.
[SAFE]: No evidence of hardcoded credentials, sensitive file access, or unauthorized network operations was detected. All site modifications require a connected Webflow Designer instance and explicit user approval for each fix.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted data (Webflow element attributes and text content) to perform its audit.
Ingestion points: Site and element data retrieved via element_tool (action get_all_elements) and data_pages_tool.
Boundary markers: The skill relies on structured analysis (Phase 2 and 3) to process element data, though no explicit delimiters are defined for the element content itself.
Capability inventory: The agent can navigate pages (de_page_tool) and modify site attributes (element_tool).
Sanitization: Not explicitly implemented in the skill instructions; however, the risk is mitigated by the narrow scope of the auditing logic and the requirement for user confirmation before any write operations. This finding is considered SAFE given the context and intended utility of the skill.

webflow-mcp:accessibility-audit