@genex-ai/cli-demo 1.35.0-dev.746 → 1.35.1-dev.748

package/dist/index.js

@@ -23564,8 +23564,9 @@ async function reportModels(apiUrl, token, opts, log, toolsOnly = false) {
   );
   for (const m of shown) {
     const plan = m.personalPlan ? `  \xB7  personal plan: ${m.personalPlan}` : "";
+    const kind = m.kind === "typesafe" ? "  \xB7  judge (classifier): judge() under a budget, not generate() or bench" : "";
     log.plain(
-      `    ${c.cyan(m.id)}  ${m.label} \u2014 ${usdPerMillion(m.inputUsdPerMillion)} in / ${usdPerMillion(m.outputUsdPerMillion)} out per million tokens${plan}`
+      `    ${c.cyan(m.id)}  ${m.label} \u2014 ${usdPerMillion(m.inputUsdPerMillion)} in / ${usdPerMillion(m.outputUsdPerMillion)} out per million tokens${plan}${kind}`
     );
   }
   if (hiddenCount > 0) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@genex-ai/cli-demo",
-  "version": "1.35.0-dev.746",
+  "version": "1.35.1-dev.748",
   "description": "Set up your project's agent workspace (.claude/.codex/.cursor in the game folder), authorize, create a game project, generate AI assets, and publish (genex CLI).",
   "type": "module",
   "bin": {

package/templates/skills/genex-llm-in-games/SKILL.md

@@ -25,6 +25,17 @@ Two modes. Picking the wrong one is the most expensive mistake on this lane:
 A popup per NPC turn is not a feature, it is an interruption. More than a call
 or two per session means a grant.
 
+**Deciding, not writing? Use the judge.** When the game needs to KNOW something
+— is this NPC lying, which of five moods is the player in, how threatening is
+the scene on a scale — and not to produce words, ask the grant's classifier with
+`judge()` instead of asking `generate()` for a label. It answers typed questions
+(`choice`, `score`, `noul`) with calibrated probabilities in about a third of a
+second, writes no prose and cannot drift out of your options. It is grant-only
+and billed as used: each call is a small fraction of a coin and accrues on the
+budget, so a check every turn is affordable where a generate every turn is not.
+`generate()` for words, `judge()` for decisions; a feature often wants both —
+the judge picks the branch, generate writes the line.
+
 ## Step 0 — is the lane live on this stand?
 
 ```bash
@@ -67,9 +78,17 @@ identity must be resolved first — `$genex-threejs-embed-auth`.
   (`billingStatus`, `reservedCoins`, `chargedCoins`, their display-USD twins).
 - `requestSpendGrant({ models, perCallMaxCoins, perCallEstimateCoins,
   disclosure: { periodLabel, estimatedCallsPerPeriod, estimatedCoinsPerPeriod },
-  maxConcurrent?, maxCallsPerMinute?, allowExternal?, idempotencyKey? })` →
+  maxConcurrent?, maxCallsPerMinute?, allowExternal?, judge?: { enabled,
+  estimatedCallsPerPeriod }, idempotencyKey? })` →
   `{ status, grantId, … the limits the player approved }`; status is `active` |
   `canceled` | `expired` | `failed` | `pending`.
+- `judge({ grantId, state, questions, idempotencyKey?, timeoutMs? })` →
+  `{ status, generationId, answers, error, billing }` — the grant's classifier,
+  answered in the same response. `questions` is 1–64 named `choice`
+  (`{ instructions, criteria: { option: description } }`), `score`
+  (`{ instructions, criteria: [levels, lowest first] }`) or `noul`
+  (`{ instructions }`); each answer has its question's type, and `noul` is a
+  probability — pick your own threshold. Needs a budget requested with `judge`.
 - `getSpendGrant(grantId)` — live state and counters; the ONE source for an
   in-game budget readout.
 - `stopSpendGrant(grantId)` — the game's own stop door. Prospective: no further
@@ -188,6 +207,15 @@ arithmetic into `DESIGN.md` beside the feature. The player sees your estimate
 attributed to the game, beside the platform's own worst case; an estimate that
 is transparently low is a grant that dies mid-session.
 
+**Using the judge too?** Add `judge: { enabled: true, estimatedCallsPerPeriod }`
+to the same request — the checks you really expect per `periodLabel`, from the
+same loop arithmetic. The sheet tells the player the game also uses a fast
+classifier billed as used, with the platform's own worst case beside your count.
+A judge-only feature passes `models: []`. A budget with the judge is coin-funded
+only, and judge calls share the budget's `maxConcurrent` and
+`maxCallsPerMinute`: batch questions into one call (up to 64) rather than one
+call per question.
+
 **Then keep the burn low, because you wrote the loop:** batch those five NPCs
 into ONE call returning five decisions, cache a decision until the situation
 that caused it changes, pick the cheapest model that passes your own check, and
@@ -285,6 +313,7 @@ These are source contracts, not a claim that every stand runs this lane —
 - [ ] A bench refused as `generation_limit` was answered with `npx genex llm status`, never a retry
 - [ ] `generate()` / `requestSpendGrant()` is the first statement of a click handler
 - [ ] A repeated-call feature uses a grant; a one-off uses `generate()`
+- [ ] A decision (a label, a yes/no, a level) is a `judge()` question under the grant, not a `generate()` asked for a word
 - [ ] Disclosure numbers derive from the real loop and are written in `DESIGN.md`
 - [ ] Calls are batched and cached; nothing fires on an invisible timer
 - [ ] Every grant-ending code has in-fiction copy and a playable fallback
@@ -316,6 +345,15 @@ cause rather than showing it for both.
 **`grant_price_unreasonable`** — the declared per-call price is far above what
 that prompt can cost on that model. Re-benchmark and declare what it prints.
 
+**`judge_not_enabled`** — the budget was approved without the judge. Request a
+new one with `judge` from the next deliberate click; never auto-renew.
+**`judge_requires_grant`** — the judge's model was sent to `generate()` with no
+budget; the judge is grant-only and is called with `judge()`.
+**`judge_period_limit`** — the checks reached the ceiling the player approved for
+the period (your own `estimatedCallsPerPeriod` at the largest size): wait, and
+if it keeps happening your declared rate is too low — fix the loop or the
+estimate, never retry in a tight loop.
+
 **`generation_limit`** — three ad-hoc calls are already in flight for this
 account, or recently stopped with their bill still pending; a pending call
 stops counting ten minutes after it was dispatched. From the bench: run