webflow-llms-txt
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external Webflow sites (sitemaps, CMS items, and page content) which could theoretically contain malicious instructions.
- Ingestion points: Site content is retrieved via
scripts/check_urls.py,references/sitemap-fallback.md, andreferences/data-api-inventory.md. - Boundary markers: Instructions in
SKILL.mdandreferences/writing-descriptions.mdexplicitly direct the agent to avoid verbatim copying of meta descriptions and to ensure every claim is evidenced on the page. - Capability inventory: The skill is limited to local file writes (
llms.txt,descriptions.md, and changelogs) and network GET/HEAD requests for auditing purposes. - Sanitization: The skill includes a dedicated utility,
scripts/search_references.py, which implements asanitize_bodyfunction using regex to proactively filter common prompt injection patterns (e.g., "ignore previous", "you are") from processed content. - [COMMAND_EXECUTION]: The skill utilizes Python scripts for file validation, URL checking, and diffing. These scripts use standard libraries and are limited to the skill's operational scope (auditing local files and target site URLs).
- [EXTERNAL_DOWNLOADS]: The skill performs network requests to the target Webflow site being audited to fetch sitemaps and verify page availability. These operations are necessary for the skill's primary purpose and do not target untrusted third-party code repositories.
Audit Metadata