What problem does it solve? Coordinating multiple AI models often leads to duplicated work, silent model swaps, unverified results, and accidental data leaks. This Skill turns the current session into a controller that decides when cross-model work is worthwhile, writes complete task packages, enforces isolation and permission boundaries, and independently verifies every external result before reporting completion. ## Core Features & Use Cases - Five routing modes: auto, fast (Grok executes well-specified work), review (a different vendor's model audits output read-only), race (two blind channels compete on identical task packages in separate workspaces), and scout (isolated Grok 4.5 search of X, Reddit, and the web with retained, verifiable evidence). - Hardened scout pipeline: The bundled run_search.py script verifies the Grok binary, sandboxes execution, validates structured results, checks Reddit dates against old.reddit.com, and retains runs for follow-up questions without re-searching. - Data egress gates: Task packages classify materials as public, internal, or sensitive; private repositories are never sent to external vendors or smuggled through temp directories. - Use Case: Ask whether a new AI coding tool is worth trying—the Skill runs a deep scout over the last 7 days of X posts, verifies key sources, and reports complaints and limitations with links, without touching your account. ## Quick Start Ask the assistant to use /ray-multimodel scout to research what X users are saying about a specific product in the last 7 days and verify the key sources before summarizing.
