Prepare CUA package releases

packages/agent/CHANGELOG.md

@@ -1,5 +1,24 @@
 # Changelog
 
+## 0.6.0 - 2026-07-10
+
+Adds explicit request-recovery and context-management policies while keeping
+provider retries and exact-empty recovery disabled by default.
+
+- `retry` adds opt-in transient provider-request retries to `CuaAgent` and
+  `CuaAgentHarness`, with configurable attempt and backoff limits. Failed
+  partial streams are buffered and discarded before a clean retry is exposed.
+- `toolResultImageReplayLimit` limits each model request to the newest four
+  tool-result images by default. It operates on a request-time projection and
+  leaves agent state and persisted sessions unchanged. Harness context hooks
+  settle before this limit is applied at the `Models` boundary.
+- `responseThreading` replaces the process-wide environment switch with a
+  constructor option for OpenAI and Tzafon `previous_response_id` chaining.
+- `emptyResponseRecovery` optionally follows a successful exact-empty response
+  with a bounded, caller-supplied pi `followUp()` message. Omitting it preserves
+  pi's normal completion behavior.
+- Updated `@onkernel/cua-ai` to 0.6.0.
+
 ## 0.5.0 - 2026-07-09
 
 Adds the browser action plane and runtime mode switching. Breaking: the

packages/agent/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@onkernel/cua-agent",
-  "version": "0.5.0",
+  "version": "0.6.0",
   "description": "Kernel browser computer-use Agent and AgentHarness classes built on pi-agent-core",
   "license": "MIT",
   "type": "module",
@@ -42,7 +42,7 @@
   "dependencies": {
     "@earendil-works/pi-agent-core": "0.80.3",
     "@earendil-works/pi-ai": "0.80.3",
-    "@onkernel/cua-ai": "0.5.0",
+    "@onkernel/cua-ai": "0.6.0",
     "@onkernel/sdk": "0.49.0",
     "sharp": "^0.34.5"
   },

packages/ai/CHANGELOG.md

@@ -1,5 +1,18 @@
 # Changelog
 
+## 0.6.0 - 2026-07-10
+
+Breaking: response threading is now configured per request instead of through
+`CUA_DISABLE_RESPONSE_THREADING`.
+
+- `CuaSimpleStreamOptions` now includes `disableResponseThreading`, allowing
+  OpenAI and Tzafon Responses API calls to send the complete current context
+  instead of continuing through `previous_response_id`.
+- Removed the process-wide `CUA_DISABLE_RESPONSE_THREADING` environment
+  variable. `@onkernel/cua-agent` users can set `responseThreading: false` on
+  `CuaAgent` or `CuaAgentHarness`; lower-level callers can pass
+  `disableResponseThreading: true` in stream options.
+
 ## 0.5.0 - 2026-07-09
 
 Introduces action planes (modes) and Anthropic native computer-use tools.

packages/ai/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@onkernel/cua-ai",
-  "version": "0.5.0",
+  "version": "0.6.0",
   "description": "Kernel-curated computer-use model access built on pi-ai",
   "license": "MIT",
   "type": "module",

packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@onkernel/cua-cli",
-  "version": "0.3.1",
+  "version": "0.4.0",
   "description": "Kernel-cloud-browser computer-use TUI built on @onkernel/cua-agent and pi-tui",
   "license": "MIT",
   "type": "module",
@@ -36,8 +36,8 @@
   "dependencies": {
     "@earendil-works/pi-coding-agent": "0.80.3",
     "@earendil-works/pi-tui": "0.80.3",
-    "@onkernel/cua-agent": "0.5.0",
-    "@onkernel/cua-ai": "0.5.0",
+    "@onkernel/cua-agent": "0.6.0",
+    "@onkernel/cua-ai": "0.6.0",
     "@onkernel/sdk": "0.49.0"
   },
   "devDependencies": {