opencode

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

• Third-party content exposure detected (medium risk: 0.30). The required runtime workflow lets the user/attacker supply arbitrary prompt text via the opencode run command and iterative process(action="submit", ... data="..."), which OpenCode then ingests as free text for the coding agent.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill includes a runtime git clone of an external repository that is fetched and then used as the agent's review context, which can directly control prompts/instructions: https://github.com/user/repo.git