browser-extract

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

• Third-party content exposure detected (high risk: 0.90). This skill explicitly opens and extracts data from rendered web pages (via browser-record/browser_snapshot/browser_eval and applying templates to arbitrary host URLs as described in SKILL.md), meaning it ingests untrusted public third-party content that is returned to the model and can influence subsequent actions despite AIDefence gates.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill runs "npx -y @claude-flow/cli@latest" at runtime (which fetches and executes code from the npm registry, e.g. https://registry.npmjs.org/@claude-flow/cli) to retrieve/store templates that the agent relies on, so remote code is executed and can directly affect agent behavior.