opencode-improved

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md workflow (Phase 1: Resource Acquisition) explicitly instructs the agent to fetch and analyze external GitHub repositories, documentation URLs, and articles (e.g., "For GitHub: Fetch relevant config files", "Documentation URL → Extract key concepts", and examples like "/improve https://github.com/..."), which are untrusted third‑party sources whose content the agent reads and uses to drive configuration decisions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches and analyzes external repositories/URLs at runtime (e.g., https://github.com/example/config), using the fetched content to drive the agent's prompts/instructions, so remote content can directly influence agent behavior.