dbear-mihomo-ai-failover
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEPERSISTENCEPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data during its network discovery and feedback workflows.
- Ingestion points: discover_provider_domains and record_web_feedback tools in SKILL.md.
- Boundary markers: Instructions in references/provider-adaptation.md require the agent to corroborate temporal browser hosts and treat them as unproven until reviewed by a user.
- Capability inventory: The skill can write configuration overlays (apply_provider_overlay) and manage system services (install_failover, start_monitor, stop_monitor) as documented in SKILL.md.
- Sanitization: SKILL.md directs the use of sanitized Provider hostnames and the omission of sensitive process or transport metadata.
- [PERSISTENCE]: The skill manages a background monitor service to ensure continuous proxy failover.
- Evidence: The instructions in references/provider-adaptation.md describe managing a LaunchAgent to ensure the monitor service remains active across sessions.
- [PRIVILEGE_ESCALATION]: The skill includes tools for installing and managing local system services.
- Evidence: The install_failover, start_monitor, and stop_monitor tools in SKILL.md allow the agent to manage the lifecycle of a daemon process on the host system, subject to explicit user authorization tokens.
Audit Metadata