@visual-z/xeno 0.9.1 → 0.10.0

package/CHANGELOG.md

@@ -1,5 +1,35 @@
 # xeno changelog
 
+## 0.10.0 — 2026-10-15
+
+Goals keep working until the evidence is ready:
+- `/goal <objective>` starts immediately. The agent can explore, refine its stages and repair failed checks without asking you to approve each stage. `/goal draft` keeps the explicit checklist workflow.
+- Checks belong to the exact candidate they tested. Changing the source, outputs or checker makes old evidence stale; resuming a session keeps completed checks and their results.
+- A ready candidate is checked before the next model request, so the agent can read counter-check results without having to end its reply first.
+- Long runs retain complete check output, recover from stalled responses and share one display budget across a batch of tool results. The sidebar shows retry reasons and separates delivery from verification.
+
+The transcript is easier to follow:
+- Tool calls show their actual running, waiting, approval and finished states, with readable command and diff previews.
+- A continuous rail connects each prompt, reply and tool result. Motion stays local to active work.
+- Colours follow your terminal by default, including light and transparent themes. Explicit theme settings still take priority, and generated text colours keep readable contrast.
+- Terminal program status and the interface use the same activity facts. OpenTUI's native OSC 7501 support reports working, blocked and done states.
+
+Optional observability:
+- Export session logs, traces and metrics to an OpenTelemetry Collector over OTLP/HTTP. Export is off by default and follows your user settings; project settings cannot enable it.
+- Exports redact credentials and omit response and command contents. Prompt contents are omitted unless you explicitly enable prompt logging.
+
+## 0.9.2 — 2026-10-14
+
+New models get a context window:
+- A model that a gateway lists without limits now uses the limits the model catalog gives for the same model under another provider. The vendor's own entry is preferred. In the sidebar the window reads `1.0m via google` instead of `window unknown`. The context budget and the output limit use the same numbers.
+- If no catalog entry has the model, the window stays unknown. xeno does not guess a default.
+- A `contextWindow` you set for the model still wins.
+- Connecting a provider no longer copies catalog numbers into `models.json`. Models you connected earlier with an empty entry are fixed without reconnecting, and they follow catalog refreshes.
+- When an Anthropic-compatible endpoint's model list states a model's input and output token limits, image input or thinking support, xeno now keeps them.
+
+Terminal integration:
+- While a session is open, xeno keeps that session's metadata file open. A terminal multiplexer can use this to tell which xeno session is running in a pane.
+
 ## 0.9.1 — 2026-10-13
 
 Replies render as Markdown again:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@visual-z/xeno",
-  "version": "0.9.1",
+  "version": "0.10.0",
   "license": "UNLICENSED",
   "description": "Segmented append-only agent log with pure context and handoff projections",
   "bin": {
@@ -10,14 +10,14 @@
     "postinstall": "node install.cjs"
   },
   "optionalDependencies": {
-    "@visual-z/xeno-darwin-arm64": "0.9.1",
-    "@visual-z/xeno-darwin-x64": "0.9.1",
-    "@visual-z/xeno-linux-arm64": "0.9.1",
-    "@visual-z/xeno-linux-x64": "0.9.1",
-    "@visual-z/xeno-linux-arm64-musl": "0.9.1",
-    "@visual-z/xeno-linux-x64-musl": "0.9.1",
-    "@visual-z/xeno-win32-arm64": "0.9.1",
-    "@visual-z/xeno-win32-x64": "0.9.1"
+    "@visual-z/xeno-darwin-arm64": "0.10.0",
+    "@visual-z/xeno-darwin-x64": "0.10.0",
+    "@visual-z/xeno-linux-arm64": "0.10.0",
+    "@visual-z/xeno-linux-x64": "0.10.0",
+    "@visual-z/xeno-linux-arm64-musl": "0.10.0",
+    "@visual-z/xeno-linux-x64-musl": "0.10.0",
+    "@visual-z/xeno-win32-arm64": "0.10.0",
+    "@visual-z/xeno-win32-x64": "0.10.0"
   },
   "files": [
     "bin/xeno.exe",