cloudflare-vectorize

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime RAG workflow (templates/rag-chat.ts / POST /chat and /chat/stream) takes outsider-authored free text from Vectorize-stored document metadata (e.g., metadata.content / metadata.title returned by VECTORIZE_INDEX.query) and injects it into the LLM prompt as Context: ${context} / Answer based on context:\n\n${context}; those stored documents are not authored by the operating user by default and can contain arbitrary text.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The document-ingestion template performs a runtime fetch of an arbitrary external URL via fetch(body.url) (templates/document-ingestion.ts, /ingest/url), and that fetched HTML/text can be ingested and later inserted into LLM prompts (e.g., used as metadata.content in templates/rag-chat.ts and passed into env.AI.run), so remote content fetched at runtime can directly control prompts.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)