@logictan/dsh-desktop-agent 0.1.7 → 0.1.8

package/README.md

@@ -53,6 +53,11 @@ Only elements carrying both a token and a non-empty label become anchors.
 None beyond installing this plugin. It publishes the `cua_driver_native__*` tools
 itself, from the Cua Driver native SDK it depends on, and dispatches to them.
 
+Those tools are registered but **not announced**: their 56 schemas are 109,857
+characters of the first request, and the work goes through `desktop_agent` instead.
+The registry resolves a call from its own layers rather than from the prompt, so a
+call to one of these names still executes — it just is not advertised.
+
 A profile that still mounts the old
 `@deepseek-ai/dsh-experimental-computer-use-cua-driver-native` provider must drop
 that row and its `@deepseek-ai/dsh-computer-use` dependency first: both would
@@ -81,11 +86,8 @@ it uses the frontmost window.
 
 ### When to call it
 
-For any desktop task that takes more than about two steps, call `desktop_agent`
-rather than driving the `cua_driver_native__*` tools yourself. The loop's ~40
-captures and decisions stay inside this one call; driving the raw tools by hand
-puts every screenshot into the agent's own context instead. The raw tools remain
-the right choice for resolving a window, or for a single action already decided on.
+For any desktop task, call `desktop_agent`. The loop's ~40 captures and decisions
+stay inside this one call rather than entering the agent's own context.
 
 For a task that lives in a web page, `browser_agent` is the better tool — it
 drives the DOM over CDP and does not depend on reading pixels. This tool is for

package/lib/driver.js

@@ -161,6 +161,21 @@ async function publishTools(ctx, driver, lifetime, pending) {
     }));
   }
 
+  // These schemas are what makes the prompt expensive and the model no better:
+  // measured, they are 109,857 characters of a 220,021-character first request,
+  // and the 1,864-session log holds 509 model-direct calls against them, spread
+  // over 27 sessions — 25 of which never used `desktop_agent` at all. The work
+  // goes through `desktop_agent`, whose sub-dispatches carry a `parent` token
+  // and are resolved from the registry's own layers, never from the prompt. So
+  // the announcement is withdrawn while the definitions stay registered and
+  // executable: `resolveExecution` consults the layer visibility, not the
+  // assembly, so a model that names one anyway still gets it executed.
+  ctx.on('system-prompt/assemble', async (_assembly, _context, next) => {
+    const assembly = await next();
+    const kept = assembly.tools.filter((tool) => !names.has(tool.name));
+    return kept.length === assembly.tools.length ? assembly : { ...assembly, tools: kept };
+  });
+
   const systemPrompt = ctx.get('systemPrompt');
   if (systemPrompt === null || systemPrompt === undefined) {
     ctx.logger?.warn?.('desktop-agent: the systemPrompt service is unavailable; the Cua Driver guidance was not registered.');

package/lib/index.js

@@ -154,10 +154,8 @@ export async function apply(ctx, config, loadSdk) {
       'asks the configured model what to do, and performs the click, key, or text entry it chose. ' +
       'Works on ordinary applications and on games that expose no usable accessibility tree, because it reads the ' +
       'picture rather than the element table. ' +
-      '\n\nPREFER THIS over calling the cua_driver_native__* tools yourself for any task that takes more than ' +
-      'about two steps. It runs the whole loop inside one call, so its 40 screenshots and decisions never enter ' +
-      'this conversation; driving those tools by hand puts every screenshot into the context instead. Reach for the ' +
-      'raw tools only to resolve a window, or for a single action you have already decided on. ' +
+      '\n\nPREFER THIS for any desktop task. It runs the whole loop inside one call, so its 40 screenshots ' +
+      'and decisions never enter this conversation. ' +
       '\n\nChoose this for native windows and desktop applications. For a task that lives in a web page, ' +
       'browser_agent is the better tool: it drives the DOM over CDP and is far more precise than a screenshot. ' +
       'Note that a covered window cannot be driven, and a game that filters injected input needs foreground ' +

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@logictan/dsh-desktop-agent",
-  "version": "0.1.7",
+  "version": "0.1.8",
   "description": "DSH desktop agent: a native-vision decision loop that drives the host desktop through the Cua Driver tools. DSH 桌面 Agent 子插件。",
   "type": "module",
   "main": "lib/index.js",