osmo-user

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime path for log summarization (osmo workflow logs ... → subagent logs-reader) ingests workflow log text produced by the user’s submitted job/container, which is effectively outsider-authored free text (not authored by the operating user) and is then placed into the agent’s LLM context for summarization.

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

• Potentially malicious external URL detected (high risk: 0.90). The cookbook-fetching reference explicitly instructs the agent to fetch remote workflow README/YAML from https://raw.githubusercontent.com/NVIDIA/OSMO/main/cookbook/README.md (and derived raw GitHub URLs) at runtime and then adapt/submit that YAML so the fetched content directly controls what code/workflows are executed, making this a runtime external dependency.