@plumbus/ai-decision-laya 0.2.1 → 0.2.2

package/CHANGELOG.md

@@ -1,6 +1,8 @@
 # Changelog
 
-## 0.2.1 — Unreleased
+## 0.2.2 — Unreleased
+
+- Document core 0.7.4+ classification with per-call provider/model and probability threshold. Link the packaged classification recipe from agent instructions and correct stale package-only guidance. Core peer compatibility remains `0.7.x`.
 
 - Run offline Laya tests with one worker and serial files. One-off smoke runs now stop servers they start on both success and failure; explicit persistent servers and the model cache are preserved.
 

package/README.md

@@ -1,10 +1,10 @@
 # @plumbus/ai-decision-laya
 
-Release notes: [changelog](./CHANGELOG.md) and [version/upgrade guide](../../docs/upgrading-voice-and-decision-release.md).
+Release notes: [changelog](./CHANGELOG.md) and [version/upgrade guide](../../docs/upgrading-classification.md).
 
 Laya provider for self-hosted typed decision inference, with a persistent Python reference service.
 
-Version `0.2.1`; required peer `@plumbus/core` exactly `0.7.x`; Node.js 20.6+.
+Version `0.2.2`; required peer `@plumbus/core` exactly `0.7.x`; Node.js 20.6+.
 Install explicitly with `pnpm add @plumbus/ai-decision-laya`. Provider packages install
 `@plumbus/ai-decision` transitively. There is no dependency on the other provider.
 
@@ -18,6 +18,11 @@ from `app/server.ts` for API and worker processes. See the
 Direct adapter calls remain useful for infrastructure tests and do not record costs
 in core. Keep these adapters separate from the text-generation provider registry.
 
+Core **0.7.4+** also supports `ctx.ai.classify({ text, labels, provider, model, threshold })`
+for generative or decision models. Start with
+`node_modules/@plumbus/core/instructions/ai-classification.md` for registration,
+model defaults, multi-label semantics, and cost recording.
+
 ## Quick start (infrastructure / smoke testing)
 
 ```ts