What problem does it solve? When a user names a capability but not a specific tool, agents waste effort guessing which AMD Instinct webtool (llm-benchmark, dc-planner, cluster-manager, demo-portal, knowledge-exchange, slide-presenter) should handle the request. This Skill maps user intent to the correct tool and its dedicated skill file, and coordinates cross-tool workflows on a local node without browser GUIs. ## Core Features & Use Cases - Intent-to-tool routing: A lookup table maps requests like GPU benchmark sweeps, DC/GPU BOM sizing, demo readiness, or slide search to the right tool and its SKILL.md entry point. - Ecosystem topology: Documents the sibling directory layout under ~/workspace, the shared/ symlink health check, default ports, and the connected-tool graph (e.g., llm-benchmark querying knowledge-exchange via /api/ask). - Cross-tool workflows: Chains outputs between tools, such as benchmark results flowing into slide-presenter decks or cluster-manager reports feeding knowledge-exchange enablement content. - Agent layer preference: Orders integration fallbacks from MCP stdio, to HTTP /mcp, to REST /api/*, to browser-level DOM automation as a last resort. - Use Case: A user asks to "run a GPU benchmark sweep and turn the results into a report deck." The Skill routes the sweep to llm-benchmark, then hands the report output to slide-presenter or knowledge-exchange for deck generation. ## Quick Start Ask the agent which webtool should handle your task, for example by saying you want to size a GPU cluster BOM or check demo readiness, and let it route the work to the right tool.
