webflow-schema
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, which could lead to indirect prompt injection if the agent follows instructions hidden within the processed content.
- Ingestion points:
scripts/audit_jsonld.pyfetches and parses HTML content from arbitrary URLs;scripts/check_bindable_text.pyreads data from user-supplied CSV files. - Boundary markers: Absent. Data from external sources is printed to the console without delimiters or instructions to the agent to ignore embedded commands.
- Capability inventory:
scripts/audit_jsonld.pyperforms network requests;scripts/check_bindable_text.pyandscripts/search_references.pyperform file system reads. - Sanitization: While
scripts/search_references.pyincludes a filter (_INJECTION_RE) for common injection patterns, the primary data processing scripts (audit_jsonld.pyandcheck_bindable_text.py) do not sanitize content before outputting it to the agent's context. - [DATA_EXFILTRATION]: The script
scripts/audit_jsonld.pyuses theurllib.request.urlopenfunction to retrieve content from user-specified URLs. This performs network operations to non-whitelisted domains. While functional, it could also be misused to access internal resources or read local system files via thefile://protocol if the agent is directed to access local paths.
Audit Metadata