Fable 5 Model Routing

Entry Gate

Before applying this skill:

1. Identify the executing harness. In Codex, do not apply this skill or use it as routing policy; continue with Codex's native capabilities. Explicit invocation does not override this exclusion.
2. In Claude Code or Cursor, inspect the user's actual goal, available context, relevant artifacts, stakes, ambiguity, and missing evidence before deciding whether the skill fits. Resolve discoverable facts rather than inferring a judgment problem from trigger words alone.
3. Apply the skill only when Fable 5 is available and the remaining problem has a consequential judgment core—such as a material trade-off, hidden premise, conflicting evidence, or defensible recommendation—for which Fable 5 has a concrete advantage over direct completion by the current agent.
4. A long task, a matching phrase, or an opportunity to use multiple agents is not enough by itself.

The fit decision is the lead's inspection, not another user confirmation. If the request already names the decision and grants its scope, resolve the eligible route and proceed. Ask only when a missing user-owned trade-off changes the recommendation; continue collecting independent facts meanwhile.

This gate controls the harness that leads the workflow. It does not prevent an eligible Claude Code or Cursor run from assigning bounded support work to a configured Codex or GPT-5.6 Sol lane after the judgment boundary is stable.

Purpose

After the entry gate passes, use Fable 5 as the lead for work whose value comes from judgment, ambiguity resolution, sustained context, or a defensible recommendation. Fable 5 can also perform difficult end-to-end work; do not delegate merely to keep its context empty. Delegate when concurrency, context isolation, fresh verification, tool specialization, or a measured cost/latency advantage improves the result.