delega-efficientelisted

# Delega efficiente Operating policy (user directive): the top model plans, judges and verifies; execution goes to the cheapest adequate means. Scripts cost zero — prefer them over any model call for deterministic work. Quality-sensitive code always goes to the top model, accepting extra tokens. Policy complexity budget (meta-rule): every addition to this skill must delete an existing policy or replace two exceptions — the skill never grows net. Same principle as the playbook's 30-line cap: each rule is a permanent cognitive cost paid at every invocation. ## Route by task properties Any route involving delegation/orchestration declares a one-line pre-budget (see "Falsifiable pre-budget"). Before planning batch/workflow/multi-agent work, consult the playbook at `~/.claude/delega-playbook.md` (init from the plugin's `playbook-template.md` if missing; a team may point it to a shared repo file via symlink). Route the REMAINDER, not the task: sunk tokens never justify staying on the current route — compare the cost of the NEXT step only. A mid-task route change is a `reversal` (log it — see Telemetry): not an error, a falsified initial decision. Score the task on six axes, top to bottom; earlier axes override later ones: 1. **Interactivity** — needs live iteration (visual, browser, user feedback)? → top model inline, never delegate. 2. **Error cost** — production code, client-facing numbers/wording, irreversible or externally visible writes? → top model, accepting extra tok