website-intelligence

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt explicitly asks the user for their Firecrawl API key and shows/requests inserting it verbatim into a MCP config JSON (and to add it to settings), which requires the LLM to handle and output the secret value directly, creating an exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested via Firecrawl at runtime when the skill scrapes the client’s and competitors’ webpages (e.g., mcp__firecrawl__scrape / mcp__firecrawl__search / mcp__firecrawl__map), which returns arbitrary page text that can include indirect prompt-injection content.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill requires a Firecrawl MCP that is started with an npx command ("npx -y firecrawl-mcp") — which will fetch and execute remote npm package code at runtime (and the doc also directs users to https://firecrawl.dev for the API key), so there is a runtime external dependency that executes remote code.