From 578ed0608ebcb6b0ad6acace86e4c58aa353ad13 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Wed, 8 Jul 2026 20:27:40 +0000
Subject: [PATCH] =?UTF-8?q?=F0=9F=A6=8B=20New=20version=20release?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .changeset/py-max-tokens-model-ceiling-clamp.md | 16 ----------------
 typescript/core/CHANGELOG.md                    | 17 +++++++++++++++++
 typescript/core/package.json                    |  2 +-
 3 files changed, 18 insertions(+), 17 deletions(-)
 delete mode 100644 .changeset/py-max-tokens-model-ceiling-clamp.md

diff --git a/.changeset/py-max-tokens-model-ceiling-clamp.md b/.changeset/py-max-tokens-model-ceiling-clamp.md
deleted file mode 100644
index 9e320bc..0000000
--- a/.changeset/py-max-tokens-model-ceiling-clamp.md
+++ /dev/null
@@ -1,16 +0,0 @@
----
-"@smooai/smooth-operator-core": patch
----
-
-feat(llm,python): clamp `max_tokens` to the model's output ceiling (Python parity)
-
-Python parity for the Rust `LlmClient` output-ceiling clamp. `AgentOptions` gains a
-`model_max_output: int | None` field and the engine now sends
-`effective_max_tokens(max_tokens, model_max_output)` =
-`min(max_tokens, ceiling)` on both the streaming and non-streaming chat paths (`None`
-/ non-positive ceiling ⇒ graceful passthrough, no behaviour change). A new
-`effective_max_tokens` helper is exported for consumers. This stops a policy/budget
-`max_tokens` from exceeding what a model can physically emit — which otherwise makes
-a reasoning model burn its budget and return empty, or 400s upstream (e.g.
-`groq-compound`'s 8192 output cap). The ceiling is sourced from the gateway's
-`/model/info` by the consumer (the server), kept out of the engine. EPIC th-1cc9fa.
diff --git a/typescript/core/CHANGELOG.md b/typescript/core/CHANGELOG.md
index fc20831..890ee30 100644
--- a/typescript/core/CHANGELOG.md
+++ b/typescript/core/CHANGELOG.md
@@ -1,5 +1,22 @@
 # @smooai/smooth-operator-core
 
+## 0.20.3
+
+### Patch Changes
+
+- ab24904: feat(llm,python): clamp `max_tokens` to the model's output ceiling (Python parity)
+
+  Python parity for the Rust `LlmClient` output-ceiling clamp. `AgentOptions` gains a
+  `model_max_output: int | None` field and the engine now sends
+  `effective_max_tokens(max_tokens, model_max_output)` =
+  `min(max_tokens, ceiling)` on both the streaming and non-streaming chat paths (`None`
+  / non-positive ceiling ⇒ graceful passthrough, no behaviour change). A new
+  `effective_max_tokens` helper is exported for consumers. This stops a policy/budget
+  `max_tokens` from exceeding what a model can physically emit — which otherwise makes
+  a reasoning model burn its budget and return empty, or 400s upstream (e.g.
+  `groq-compound`'s 8192 output cap). The ceiling is sourced from the gateway's
+  `/model/info` by the consumer (the server), kept out of the engine. EPIC th-1cc9fa.
+
 ## 0.20.2
 
 ### Patch Changes
diff --git a/typescript/core/package.json b/typescript/core/package.json
index 467877c..1e80a73 100644
--- a/typescript/core/package.json
+++ b/typescript/core/package.json
@@ -1,6 +1,6 @@
 {
     "name": "@smooai/smooth-operator-core",
-    "version": "0.20.2",
+    "version": "0.20.3",
     "description": "Native TypeScript implementation of the smooth-operator agent engine — an in-process, OpenAI-compatible agentic tool-calling loop with knowledge grounding. The TypeScript sibling of the Rust reference engine, the C# core, and the Python core.",
     "type": "module",
     "license": "MIT",