46-orchestratelisted

# 4.6 Orchestrate <p align="center"><img src="assets/architecture.svg" alt="46-orchestrate: a gpt-5.6-sol Codex orchestrator at xhigh effort owns hard decisions and routes bounded work to out-of-band Terra workers, and reaches a Fable 5 cross-vendor peer for a genuine decorrelated check on high-stakes calls; Luna is used only for tightly specified mechanical implementation" width="900"></p> ## Preflight the lead runtime first Before reading the task brief, inspecting the workspace, planning, or delegating, resolve `SKILL_DIR` as the directory containing this `SKILL.md` and run: ```bash "$SKILL_DIR/scripts/check-lead-runtime.sh" ``` This is fail-closed. It uses the current process's `CODEX_THREAD_ID` to locate that exact thread's rollout, reads the latest recorded `turn_context` for the current turn's model and effort, and rejects any model-reroute event recorded for that turn. Proceed only when it prints `46-orchestrate preflight OK: gpt-5.6-sol at xhigh`. If it reports a mismatch or cannot establish the runtime, stop immediately, quote the result to the operator, and ask them to select Sol with Extra High reasoning in `/model` (or restart with `--model gpt-5.6-sol -c model_reasoning_effort=xhigh`), verify with `/status`, and invoke `$46-orchestrate` again. Never infer the runtime from this skill's prose, the launch command the operator says they used, `config.toml`, or the newest session file. Act as the lead orchestrator for the GPT-5.6 family: plan, decompose, delega