planning

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

• Third-party content exposure detected (high risk: 1.00). The Research & Analysis instructions (references/research-phase.md) explicitly require reading and analyzing external GitHub repositories, PRs/issues/discussions (via gh and repomix --remote) and external documentation, so the agent ingests untrusted public third‑party content that can influence planning decisions.

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

• Potentially malicious external URL detected (high risk: 0.80). The Research phase explicitly shows a runtime command (repomix --remote https://github.com/mrgoonie/human-mcp) that fetches a GitHub repo and uses its contents to generate a codebase summary injected into the agent's context, so the remote URL is used at runtime to control prompts.