openspace

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

• Third-party content exposure detected (high risk: 0.75). The required workflow includes searching/downloading and then applying community/cloud skills (public web content and outsider-authored skill text) such that the LLM can ingest that skill’s instructions during execution (e.g., via openspace-download-skill/execute_task flows described in SKILL.md).

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

• Potentially malicious external URL detected (high risk: 0.80). The SKILL explicitly documents runtime downloads and MCP searches that fetch and apply cloud-hosted skills (which become agent instructions) from https://open-space.cloud, so external content from that URL can directly control agent behavior at runtime.