agentus 0.9.0 → 0.10.1

package/README.md

@@ -53,10 +53,11 @@ depth**, a permission mode that survives a restart, the **context-budget** field
 channel** (memory recall/save hints in the message stream), per-process skill roots, **project
 skills made visible and switchable** (a repo's `.agents/skills` shows up as a `project_skills`
 option instead of being skipped in silence), **background-job completion wake-ups** (a finished
-background process comes back as a turn of its own), and a **DeepSeek web backend** (search +
+background process comes back as a turn of its own), an **ACP session that writes its own log files** (`agent.log`/`errors.log`, so a
+cockpit-side failure is diagnosable afterwards), and a **DeepSeek web backend** (search +
 extract on the model key) — come from the fork's integration build instead. That is a five-line install:
 [§ Which Hermes to run](docs/installation.md#which-hermes-to-run) — currently the tag
-`agentus/v0.22.0/merge-20261021-1400`, on Hermes `v0.22.0`.
+`agentus/v0.22.0/merge-20261022-2226`, on Hermes `v0.22.0`.
 
 The first slot you start will use the Hermes home **your own agent already uses**
 (`~/.hermes`) unless the backend row names another one — see "HERMES_HOME" below.

package/README.zh-CN.md

@@ -47,10 +47,11 @@ web bundle、配套 APK —— 然后直接退出，不启动任何东西）。
 **上下文预算**输入框、把记忆召回 / 存记忆提示送进消息流的 **status 通道**、按进程指定的技能根、
 **项目 skill 可见可设**（仓库的 `.agents/skills` 会以 `project_skills` 选项出现，而不是被无声跳过）、
 **后台任务完成唤醒**（后台进程跑完会自己开一个回合回来），
+**ACP 会话自己写日志**（`agent.log`/`errors.log`，驾驶舱里出的问题事后可查），
 以及 **DeepSeek 的 web 后端**（搜索 + 正文抓取，共用模型 key）—— 来自 fork 的集成构建（tag）。
 装法就五行：
 [§ 该跑哪份 Hermes](docs/installation.zh-CN.md#该跑哪份-hermes) —— 当前是 tag
-`agentus/v0.22.0/merge-20261021-1400`，基座 Hermes `v0.22.0`。
+`agentus/v0.22.0/merge-20261022-2226`，基座 Hermes `v0.22.0`。
 
 你启动的第一个 slot 会使用**你自己的 agent 已经在用的**那个 Hermes home（`~/.hermes`），除非后端那一行
 指定了另一个 —— 见下面的「HERMES_HOME」。

package/docs/installation.md

@@ -60,7 +60,7 @@ ref and gets rebuilt, while a tag never moves — so pin the tag. The `agentus/`
 ours apart from upstream's own tags (`v0.22.0`, `v0.22.0+canary.*`), and the timestamp in the
 last segment sorts, so "newest" is a `sort -V`.
 
-**Current integration build: `agentus/v0.22.0/merge-20261021-1400`** — the newest tag on the fork,
+**Current integration build: `agentus/v0.22.0/merge-20261022-2226`** — the newest tag on the fork,
 built on Hermes `v0.22.0`. This line is for orientation only, not the source of truth: the
 `TAG=$(…)` step below resolves the newest one on every run, so read it there if this sentence
 ever looks stale.

package/docs/installation.zh-CN.md

@@ -52,7 +52,7 @@ agentus/<hermes 内部版本>/merge-<YYYYMMDD-HHMM>
 tag。`agentus/` 这个前缀也把它和上游自己的 tag（`v0.22.0`、`v0.22.0+canary.*`）分开；最后一段是时间戳，
 可排序，所以「最新」就是一次 `sort -V`。
 
-**当前集成构建：`agentus/v0.22.0/merge-20261021-1400`** —— fork 上最新的那个 tag，基座 Hermes
+**当前集成构建：`agentus/v0.22.0/merge-20261022-2226`** —— fork 上最新的那个 tag，基座 Hermes
 `v0.22.0`。这行只为一眼看个大概，不是权威：下面的 `TAG=$(…)` 每次都会自己解析出最新，觉得这行过期就
 以它为准。
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentus",
-  "version": "0.9.0",
+  "version": "0.10.1",
   "description": "Multi-session web cockpit for ACP-speaking coding agents (Qoder CLI / Hermes). UI has no intelligence — the agent lives in spawned CLI subprocesses.",
   "license": "Apache-2.0",
   "repository": {
@@ -78,12 +78,15 @@
     "live-stream": "tsx scripts/qa/live-stream.mts",
     "open-target": "tsx scripts/qa/open-target.mts",
     "rail-recent": "tsx scripts/qa/rail-recent.mts",
+    "rail-breakpoints": "node scripts/qa/rail-breakpoints.mjs",
     "permission-echo": "node scripts/qa/permission-echo.mjs",
     "turn-pulse": "tsx scripts/qa/turn-pulse.mts",
     "resume-tail": "tsx scripts/qa/resume-tail.mts",
     "render-lag-sweep": "node scripts/qa/render-lag-sweep.mjs",
     "replay-window": "node scripts/qa/replay-window.mjs",
-    "think-cadence": "node scripts/qa/think-cadence.mjs"
+    "think-cadence": "node scripts/qa/think-cadence.mjs",
+    "vad": "tsx scripts/qa/vad.mts",
+    "vad-silero": "tsx scripts/qa/vad-silero.mts"
   },
   "engines": {
     "node": ">=22.5"

package/packages/server/src/settings.ts

@@ -62,7 +62,7 @@ export const THEME_DEFAULT: ThemeSettings = { mode: "system", accent: "" };
  *  what a phone's own loudspeaker leaks back into its microphone is nothing like a headset
  *  in a quiet room — so the defaults are a starting point, not a truth. The call page owns
  *  the panel; the settings page reads the same numbers. */
-export type InterruptMode = "voice" | "button";
+export type InterruptMode = "voice" | "button" | "keyword";
 
 export interface CallSettings {
   /** WHO gets to interrupt the agent, and the operator picks:
@@ -85,6 +85,10 @@ export interface CallSettings {
    *  as a turn (1 = no limit). A tap on the orb always sends what is there — that gesture
    *  is an explicit instruction, not a guess about where a sentence ended. */
   minChars: number;
+  /** 关键词打断 — the phrase that must be HEARD before the floor changes hands (opt-in; empty means
+   *  the mode is inert). No shipped default on purpose: nobody should be able to interrupt this
+   *  agent by guessing a word we chose for them. */
+  wakePhrase: string;
 }
 
 /** How the operator wants to TALK to the agent. These were browser-local (localStorage) and
@@ -150,14 +154,32 @@ export const PREFS_DEFAULT: PrefsSettings = {
   planHint: { on: true, everyTurn: true, instruction: "" },
 };
 
+/** The longest wake phrase, in CODE POINTS (an emoji is one, not two). One number, one place — the
+ *  web side has its own copy (`WAKE_MAX` in callSettings.ts) and the two must keep saying 24. */
+const WAKE_MAX = 24;
+
+/** Drop half-pairs left by a client that counted UTF-16 units and cut in the middle of an emoji.
+ *  `for...of` walks code points, so a lone surrogate arrives as a one-character string and is easy
+ *  to spot; storing it would put U+FFFD in the database. */
+function stripLone(s: string): string {
+  let out = "";
+  for (const ch of s) {
+    const cp = ch.codePointAt(0) ?? 0;
+    if (cp >= 0xD800 && cp <= 0xDFFF) continue;   // a surrogate that did not pair with its other half
+    out += ch;
+  }
+  return out;
+}
+
 export const CALL_DEFAULT: CallSettings = {
   /* 说完停顿 1.5s (was 1.2): the operator reads a pause as "it stopped listening" before 1.2 s,
      and a long sentence with a breath in the middle got cut. Both ends of the range still let
      the room decide — this is only where a fresh install starts. */
   interruptMode: "voice", bargeSensitivity: 60, bargeMs: 300, silenceMs: 1500, minChars: 3,
+  wakePhrase: "",
 };
 
-export const INTERRUPT_MODES: InterruptMode[] = ["voice", "button"];
+export const INTERRUPT_MODES: InterruptMode[] = ["voice", "button", "keyword"];
 
 /** How the cockpit treats AGENT PROCESSES — as opposed to what a conversation contains. AionUi
  *  ships the same knob in 设置 → 系统 as 「Agent 空闲超时（分钟）」 (default 5, 1–60, step 5) with the
@@ -369,7 +391,11 @@ function callAtShippedDefaults(r: Record<string, unknown>): boolean {
   // The mode counts too: a row that only says `interruptMode: "button"` IS a deliberate choice and
   // must not be reset along with a section that happens to still carry the old numbers.
   const modeShipped = r.interruptMode === undefined || r.interruptMode === CALL_DEFAULT.interruptMode;
-  return sens && modeShipped && Number(r.bargeMs) === 300 && Number(r.silenceMs) === 1200 && Number(r.minChars) === 1;
+  // …and a stored wake phrase is a deliberate choice too: without this, setting one and switching
+  // back to 语音打断 would make the whole section look untouched and the phrase would be dropped.
+  const wakeShipped = r.wakePhrase === undefined || r.wakePhrase === CALL_DEFAULT.wakePhrase;
+  return sens && modeShipped && wakeShipped
+    && Number(r.bargeMs) === 300 && Number(r.silenceMs) === 1200 && Number(r.minChars) === 1;
 }
 
 /** What each knob may be. The ranges are the panel's slider ends; the server is the one
@@ -503,6 +529,12 @@ function pickCall(raw: unknown): CallSettings {
     if (callAtShippedDefaults(raw as Record<string, unknown>)) return out;
     const mode = (raw as Record<string, unknown>).interruptMode;
     if (INTERRUPT_MODES.includes(mode as InterruptMode)) out.interruptMode = mode as InterruptMode;
+    const wake = (raw as Record<string, unknown>).wakePhrase;
+    // NOT length-capped here on purpose: this is what somebody ELSE stored (curl, a script, a previous
+    // version). Cutting it on the way out would then be written straight back by the `persist()` that
+    // runs at boot — silently destroying the operator's phrase with no log and nothing on screen.
+    // Length belongs to the write path (`saveCall`), which is where a human typed it.
+    if (typeof wake === "string") out.wakePhrase = wake.replace(/\s+/g, " ").trim();
     for (const k of CALL_NUMERIC) {
       const v = Number((raw as Record<string, unknown>)[k]);
       const [lo, hi] = CALL_RANGE[k];
@@ -734,10 +766,24 @@ export function saveCall(patch: Record<string, unknown>): Settings {
   if ("interruptMode" in patch) {
     const mode = patch.interruptMode;
     if (!INTERRUPT_MODES.includes(mode as InterruptMode)) {
-      throw new Error(`invalid interruptMode: ${JSON.stringify(mode)} (expected voice|button)`);
+      throw new Error(`invalid interruptMode: ${JSON.stringify(mode)} (expected voice|button|keyword)`);
     }
     c.interruptMode = mode as InterruptMode;
   }
+  if ("wakePhrase" in patch) {
+    if (typeof patch.wakePhrase !== "string") throw new Error("invalid wakePhrase: expected a string");
+    // Truncated, not rejected: this field arrives in the same body as the sliders, and refusing it
+    // would throw away the operator's OTHER changes with it (and the panel would then say "saved
+    // locally, not on the server" for a phrase nobody was trying to abuse). Same rule as pickCall.
+    const wake = stripLone(patch.wakePhrase.replace(/\s+/g, " ").trim());   // an old client can slice a surrogate in half
+    const kept = [...wake];
+    if (kept.length > WAKE_MAX) {
+      // Truncated rather than refused, but never silently: one line in the log is the difference
+      // between 「我设过这个词」 and 「它自己变短了」 when a script writes 40 characters.
+      console.warn(`[agentus] wakePhrase truncated: ${kept.length} → ${WAKE_MAX} characters`);
+    }
+    c.wakePhrase = kept.slice(0, WAKE_MAX).join("");
+  }
   for (const key of CALL_NUMERIC) {
     if (!(key in patch)) continue;
     const raw = typeof patch[key] === "number" ? patch[key] as number : Number(patch[key]);