taste

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly sends the user-provided URL to the Playwright MCP (mcp__playwright__browser_navigate) and requires substituting the original URL verbatim into the output JSON's source field, so any secrets embedded in a URL (user:pass@host, tokens in query strings) would be included in LLM-generated output — an exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.85). Yes: the required runtime workflow navigates to a user-supplied URL and runs a DOM extractor (mcp__playwright__browser_evaluate) that reads computed styles/DOM-derived data from that outsider-authored webpage, which is then fed into the LLM for Steps 1–4.