ai-rag-pipeline

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md shows the pipeline fetching public web search results and URL content via tavily/search-assistant, exa/search and tavily/extract (e.g., SEARCH_RESULT=$(infsh app run tavily/search-assistant ...) and CONTENT=$(infsh app run tavily/extract ...)) and then injects those untrusted, user-generated third-party contents directly into LLM prompts, so external pages can materially influence agent behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly runs infsh's extraction at runtime to fetch arbitrary web pages and injects the resulting CONTENT into LLM prompts (example call: infsh app run tavily/extract with URLs such as https://example.com/research-paper), so externally-hosted content fetched at runtime can directly control the agent's prompt/context.