vllm-semantic-router-harness: Скилл для автоматизации задач

--- name: vllm-semantic-router-harness description: Bridges native skill discovery into the vLLM Semantic Router repository harness, routing tasks through the canonical agent-report flow, repo-local skill registry, and validation commands. Use when starting any task inside the vLLM Semantic Router repository to resolve the correct primary skill, read canonical docs, and run harness validation. --- # vLLM Semantic Router Harness This is a native-discovery bridge. The canonical source of truth remains: - [`AGENTS.md`](../../../AGENTS.md) - [`docs/agent/README.md`](../../../docs/agent/README.md) - [`tools/agent/skill-registry.yaml`](../../../tools/agent/skill-registry.yaml) - [`tools/make/agent.mk`](../../../tools/make/agent.mk) ## When To Use Use this skill whenever the current workspace is this repository or the task touches files under `src/`, `tools/agent/`, `docs/agent/`, `dashboard/`, or `.github/`. ## Required Flow 1. Read [`AGENTS.md`](../../../AGENTS.md) first. 2. Read [`docs/agent/README.md`](../../../docs/agent/README.md) and the nearest local `AGENTS.md` for hotspot directories. 3. Run `make agent-report ENV=cpu|amd CHANGED_FILES="..."` before non-trivial edits. 4. Start from the resolved primary skill under [`tools/agent/skills/`](../../../tools/agent/skills/). 5. Validate with the canonical harness commands from [`tools/make/agent.mk`](../../../tools/make/agent.mk). ## Boundaries - Do not treat this wrapper as canonical policy text. - Do not copy or fork skill instructions out of `tools/agent/**`. - Use `make agent-report` to expose the full routed context pack; this wrapper only improves native startup discovery.