What problem does it solve? Orchestrating multiple local AI model CLIs (Grok, Kimi Code, Claude Code) by hand means writing raw CLI flags, waiting on serial calls, and trusting exit codes that do not prove files were actually produced. This Skill routes each task to the right CLI with strong default models, runs independent jobs concurrently, verifies artifacts, and retries only failed jobs. ## Core Features & Use Cases - Model Routing: Sends X/web search, image, and video tasks to Grok CLI (grok-4.5), frontend UI work to Kimi Code CLI (kimi-code/k3, 1M context), and complex coding or review to Claude Code CLI (Fable 5 default, Opus 4.8 / Sonnet 5 optional). - Concurrent Batch with Done Gate: Launches independent jobs in one batch call with isolated working directories, native JSON stream progress events, private per-job logs, and expects/verify_commands artifact checks so a zero exit code alone never counts as success. - Retry and Cancellation: Ctrl-C terminates all active process groups, and retry --summary reruns only failed jobs from a prior run without repeating successful ones. - Use Case: Ask three models to each build a landing page version; the Skill creates one batch config, starts Grok, Kimi, and Claude concurrently in separate directories, verifies each index.html exists, and reports a side-by-side summary. ## Quick Start Ask the agent to use qiaomu-model-cli to run a task, for example: use grok cli to search the latest posts from an X account, or run a batch comparing Grok 4.5, Kimi K3, and Claude Opus 4.8 on the same page-building task.
