wizz-method 2.8.1 → 2.9.0

package/package.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
   "name": "wizz-method",
-  "version": "2.8.1",
+  "version": "2.9.0",
   "description": "Wizz Method: método de agência orientado por IA em PT-BR",
   "keywords": [
     "agile",

package/src/skills-lib/wizz-router/SKILL.md

@@ -40,6 +40,12 @@ Assess the **Area**: how many areas does the request touch? The canonical area l
 
 **Dispatch rule:** IF 2+ areas → `wizz-maestro`, always. OTHERWISE count the 3 remaining factors (multi-step, needs planning, produces a memorable artifact); 2+ factors → `wizz-maestro`. The area clause decides on its own, without combining with anything else.
 
+**One gate runs before dispatch, and it is deliberately narrow: the ambiguous referent.** Ask the user instead of dispatching ONLY when all three hold: (a) the request names a target that does not resolve to ONE thing — the context shows several objects that fit the name, or there is no antecedent at all; (b) a single short factual question settles which one; (c) guessing wrong lands the work on the wrong object at a cost worth a question. When it fires, ask via `AskUserQuestion` and dispatch only after the answer. "Update the price on the page" with three live sales pages is the case; "improve this" with nothing to point at is the case; "remove the banner" where a promo banner and a cookie banner both exist is the case.
+
+**Clause (c) is eliminatory, and it is the clause that settles most cases — check it last and let it close the gate.** The bar is that a wrong guess costs more than redoing the delivery: a change lands on the wrong live object, or work already done has to be thrown away. A short piece of text that is cheap to rewrite never clears that bar, however unresolved its referent looks: "write a headline to sell this plan" has no antecedent at all for "this plan", clause (a) holds, and it still goes to `agent:copy` — a headline aimed at the wrong plan is rewritten in a minute. **Clause (a) alone never fires the gate.** (Calibrated on `a04`, measured 2026-09-21: the gate fired there and should not have.)
+
+This gate does NOT fire for an open-scope request. "My site is bad, sort it out", "my sales page converts badly": the target resolves fine, it is the SCOPE that is undefined, and discovering scope is the Manager's work — that is the counterpart clause below, and it stays the maestro's. It does NOT fire for an area that is defensible either way: pick the area, do not ask. And it never outranks the sensitive-subsystem clause below — an ambiguous request that touches authentication, payments, permissions or production data goes to `wizz-maestro`, who asks as part of planning what must not break. One rule per case: the Director asks only about WHICH object, never about what the work should be. **When in doubt, the gate does NOT fire.** (Owner's decision, 2026-09-21.)
+
 **Decision order — capabilities → required relations → areas → shortcuts.** Match the catalogue entries, inspect their `rel`, and include only capabilities needed for the requested deliverable and not already supplied in the context. Resolve their owning areas before applying ANY one-area shortcut. One artifact can require multiple areas; the word "copy" does not prove that only copy is needed. For full sales-page copy, if either the offer or the conversion structure is missing from the supplied context, include the missing capability: `wizz-offer-forge` for the offer and/or `wizz-page-cro` for conversion structure (growth), alongside `wizz-copywriting` (copy); route to maestro. Do not assume those inputs are ready merely because the user did not mention them. If both are already approved and the request is limited to text, keep copy-only work in copy. An isolated headline or CTA does not require redesigning the whole page. Optional neighbours and separate deliverables do not add areas; `pairs_with` alone never escalates.
 
 **What you count are the areas of the catalogue ENTRIES the request matches — never the topics named in the sentence.** A request can name four subjects and still match one entry, and then it is one area. "Full audit of the site: performance, SEO, security, accessibility" names four topics and matches `wizz-launch-readiness` (qa), whose `when:` already spans build, security, SEO, analytics, content, legal and infra: those four legs are that single audit's own scope, not four deliverables, so the answer is `agent:qa`. Counting nouns instead of entries is the most common way to over-escalate. (Owner's decision, 2026-09-19.)