subagent-model-routing

Route work to non-Claude models (codex/GPT-5.x, Kimi, xAI Grok 4.5, GLM, MiniMax, and local models through opencode custom providers) over the codex/kimi/opencode/grok standalone shims. Supports both one-shot flat dispatch and dependency-ordered DAG orchestration via the Workflow tool. Use when delegating authoring, review, analysis, throughput, or multi-step work to external agentic CLI harnesses. Section 0 decides flat vs DAG; Picking the model decides which model.

Codex-native workflow for routing work to local external-model shims: Claude models through claude-shim.sh, Kimi through kimi-shim.sh, Grok 4.5 through grok-shim.sh, and GLM/MiniMax through opencode-shim.sh. Local/self-hosted models (including Qwen) route through opencode-shim.sh as a custom provider. Use when the user asks Codex to dispatch, compare, review, fan out, or sequence work across those local shims. This port avoids Claude Code orchestration surfaces and uses the shared model-routing JSON workflow runner for durable external-only dependency graphs.

Copilot-compatible workflow for routing work to local external-model shims: codex/GPT-5.x through codex-shim.sh, Claude models through claude-shim.sh, Kimi through kimi-shim.sh, Grok 4.5 through grok-shim.sh, and GLM/MiniMax through opencode-shim.sh. Local/self-hosted models (including Qwen) route through opencode-shim.sh as a custom provider. Use when the user asks Copilot to dispatch, compare, review, fan out, or sequence work across those local shims. This port avoids Claude Code orchestration surfaces and uses the shared model-routing JSON workflow runner for durable dependency graphs.