buying-signals-6

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

• Third-party content exposure detected (high risk: 0.75). The runtime workflow explicitly includes “ColdIQ MCP… mcp__coldiq__search_web for press releases and funding news” and the fallback mentions manual LinkedIn/BuiltWith/Crunchbase queries, which are outsider-authored free text (public web pages / third-party posts) that can be fetched and ingested into the agent’s LLM context via those tool calls.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill includes an explicit installer command that fetches and executes remote code ("COLDIQ_API_KEY= npx -y @coldiq/mcp@latest" at SKILL.md:13) and references the ColdIQ endpoints (https://api.coldiq.com and https://coldiq.com/marketplace) used to enable that runtime integration, so there is a clear runtime dependency that executes remote code.