hmos-multidevice-scenario-entry

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

• Third-party content exposure detected (high risk: 0.75). Required workflow includes reading scene SKILL.md and references via Read tools (and may remote-install missing skills), so the LLM ingests outsider-authored free text from remotely downloaded/public Matrix skills’ SKILL.md/references at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill mandates at-runtime auto-loading via scripts/remote_load.sh which performs a curl download from https://matrix.openharmony.cn/api/registry/skill/${SKILL_NAME}/install?format=zip (references/remote-skill-loading.md whitelist) and then the agent reads the fetched SKILL.md files into its context to drive routing/implementation, so remote content downloaded at runtime can directly control agent prompts and behavior.