wezard 2.0.4 → 2.0.5

package/dist/daemon/mirror-bridge.js

@@ -26,7 +26,7 @@ import { noticeSuffixFor } from "./notices.js";
 import { dangerOf } from "./danger.js";
 import { runTmux, spawnTmuxClaude } from "./spawn-tmux.js";
 import { selectModel } from "./model-select.js";
-import { hasRegistry, markTranscript, probeOf, sessionOnPane, submittedSince } from "./cc-session.js";
+import { hasRegistry, markTranscript, probeOf, sessionOnPane, sessionPanes, submittedSince } from "./cc-session.js";
 import { wizardStore } from "./wizard.js";
 import { startSubagentWatch } from "./subagent-tail.js";
 import { recordTool, recordToolResult, recordMark, recordTurnStart, recordTurnItem, recordTurnUsage, recordTurnClose, recordCloseOpenTurns, buildDetailUrl, buildChatUrl } from "./detail.js";
@@ -34,7 +34,7 @@ import { labelFor, tagOfKey, baseOfKey, keyOf, stripSigil, displayName, withTagH
 import { splitMarkdown } from "../shared/md-chunk.js";
 import { randomTip } from "./tips.js";
 import { chatBaseOf, chatNameOf, listChatNames, parsePeerRef, peerAddress } from "./chat-name.js";
-import { stripAnsi, compactPane, paneIsBusy, paneIsStalled, transcriptStalled, summarizeTail, lastAssistantText, lastReply, unwrapPasted, lastContextTokens, keepaliveStamps, tailTurns, renderDialog } from "./peers.js";
+import { stripAnsi, compactPane, paneIsBusy, paneIsStalled, transcriptStalled, summarizeTail, lastAssistantText, lastReply, unwrapPasted, lastContextTokens, keepaliveStamps, openKeepalivePing, tailTurns, renderDialog } from "./peers.js";
 // PATH augmentation: launchd / systemd start the daemon
 // with a stripped PATH that often lacks nvm / homebrew, breaking spawn(claudeBin).
 // Backend resolution is per-transcript, not global: `backendForPath` recovers
@@ -1565,6 +1565,9 @@ const queues = new Map();
  *  behind it forever. Well under brief's 350s bubble ceiling so the user still
  *  gets an error bubble rather than a silent stream timeout. */
 const INJECT_JOB_TIMEOUT_MS = 90_000;
+/** 没有绑定的 pane 静默多久才算孤儿。刚 spawn、还没 attach 的 pane (并发 fan-out 时
+ *  别人的 sweep 会撞见它) 在这段时间里一直有输出, 远大于一次 spawn → attach 的耗时。 */
+const ORPHAN_GRACE_MS = 10 * 60_000;
 /** Drop a session's queue chain. A hung job can't be cancelled (it's parked in a
  *  tmux read), but detaching the chain means the NEXT message doesn't inherit
  *  its deadlock. Pair with an injectGen bump so the zombie can't act when it
@@ -2830,6 +2833,7 @@ export const startMirror = (deps) => {
     // slower than the 60s fail-safe. Reuses the detail turn the timer flow may
     // have opened (fireKeepalive skips opening when this got there first).
     const KEEPALIVE_CONTENT_FAILSAFE_MS = 5 * 60_000;
+    const PONG_RE = /pong/i;
     const beginKeepaliveByContent = (a, text) => {
         a.keepaliveByContent = true;
         if (a.keepaliveContentTimer)
@@ -2883,6 +2887,19 @@ export const startMirror = (deps) => {
         // keepalive 门之前: 吞没窗口里到达也必须销账, 否则账本永久悬空。
         if (item.kind === "skill_output" && item.agentDone)
             clearBackgroundAgent(a, item.agentDone);
+        // Reply-side belt: a "pong" arriving with NO swallow holding. Both triggers
+        // above live in memory, and the tail re-attaches at EOF — a reload between
+        // the ping and its reply forgets the ping was ever sent, and the pong opens
+        // a chat turn of its own. So ask the transcript: if the last thing said to
+        // the model is a keepalive ping, this text is its pong — swallow from here
+        // to the turn_end, recorded as the same marked detail turn. Gated on the
+        // transcript, not on the word alone: a real answer that mentions pong has a
+        // real question in front of it.
+        if (item.kind === "text" && !a.keepaliveQuiet && !a.keepaliveByContent && PONG_RE.test(item.body)) {
+            const ping = openKeepalivePing(a.jsonlPath, keepalivePingSigs);
+            if (ping)
+                beginKeepaliveByContent(a, ping);
+        }
         if (a.keepaliveQuiet || a.keepaliveByContent) {
             const id = a.keepaliveTurnId;
             if (id) {
@@ -5285,18 +5302,55 @@ export const startMirror = (deps) => {
             // 一次 list-panes 同时回答「活着吗」和「有人盯着吗」—— 跑了几个月的机器上
             // allTargets 是几百个 key, 见 target 就问 tmux 会打穿 fd (同 worldPeers 的
             // WORLD_PROBE 取舍)。
-            const snap = await runTmux(["list-panes", "-a", "-F", "#{pane_id}\t#{window_active}:#{session_attached}"]);
+            const orphans = [];
+            const snap = await runTmux(["list-panes", "-a", "-F", "#{pane_id}\t#{window_active}:#{session_attached}\t#{session_name}\t#{window_activity}"]);
             if (snap.code !== 0)
-                return { reaped, skipped };
-            const watchedBy = new Map(snap.stdout.split("\n").filter(Boolean).map((l) => {
-                const [id, w = ""] = l.split("\t");
+                return { reaped, orphans, skipped };
+            const rows = snap.stdout.split("\n").filter(Boolean).map((l) => l.split("\t"));
+            const watchedBy = new Map(rows.map(([id = "", w = ""]) => {
                 const [wActive, sAttached] = w.split(":");
                 return [id, wActive === "1" && sAttached !== "0"];
             }));
-            const live = allTargets().filter((t) => watchedBy.has(paneOf(t)));
+            // 收一个 pane; 返回没收成的原因, 收掉了返回 undefined。
+            const putDown = async (pane) => {
+                // 人正盯着这个 window —— 收掉他的视野比省下那点内存更讨嫌。
+                if (watchedBy.get(pane))
+                    return "watched";
+                if (paneIsBusy(await capturePaneTail(pane, 12)))
+                    return "busy";
+                // 与 killPane 同款前戏: Esc 给生成中的 CLI 一个收束落盘的机会。
+                await runTmux(["send-keys", "-t", pane, "Escape"]);
+                await sleep(250);
+                const r = await runTmux(["kill-pane", "-t", pane]);
+                return r.code === 0 ? undefined : `kill-pane failed: ${r.stdout.slice(-120) || r.code}`;
+            };
+            const targets = allTargets();
+            // 孤儿: 住在 daemon 自己的 tmux session 里、却没有任何绑定指着它的 pane —— 绑定
+            // 丢了 pane id (restore 认不出 → 下一条消息另起新 pane, 旧 pane 不杀) 之后留下的。
+            // 没有地址能再叫到它, 却不在下面 `live` 的账上, 攒多少都触不到上限; 所以不占
+            // 名额、不等超限, 见到就收。刚起的 pane 在 attach 之前也长这样 (并发 spawn),
+            // 用 window_activity 留一段宽限。只收会话 pane: 注册表认得里面住着活的 AI 会话
+            // 才算; 普通 pane (人自己开的 shell / 编辑器, 会话已退回 shell 的, 没有注册表的
+            // 后端) 一律跳过 —— 认不出就不动。
+            const bound = new Set(targets.map(paneOf).filter(Boolean));
+            const hosting = new Set(activeBackends().flatMap((b) => [...sessionPanes(b.homeDir)]));
+            for (const [pane = "", , session, activity] of rows) {
+                if (session !== cfg.wrc.tmuxPrefix || bound.has(pane) || !hosting.has(pane))
+                    continue;
+                if (now - Number(activity) * 1000 < ORPHAN_GRACE_MS)
+                    continue;
+                const why = await putDown(pane);
+                if (why) {
+                    skipped[pane] = why;
+                    continue;
+                }
+                orphans.push(pane);
+                log.info({ pane, idleHours: Math.round((now - Number(activity) * 1000) / 3.6e6) }, "mirror pane cap — orphan pane killed (no binding)");
+            }
+            const live = targets.filter((t) => watchedBy.has(paneOf(t)));
             let over = live.length - max;
             if (over <= 0)
-                return { reaped, skipped };
+                return { reaped, orphans, skipped };
             // 超了才 stat: 最久没动的先收。还没写过 transcript 的是刚起的 pane, 按"此刻"算。
             const mtimeOf = (target) => {
                 try {
@@ -5318,21 +5372,9 @@ export const startMirror = (deps) => {
                     continue;
                 }
                 const pane = paneOf(target);
-                // 人正盯着这个 window —— 收掉他的视野比省下那点内存更讨嫌。
-                if (watchedBy.get(pane)) {
-                    skipped[target] = "watched";
-                    continue;
-                }
-                if (paneIsBusy(await capturePaneTail(pane, 12))) {
-                    skipped[target] = "busy";
-                    continue;
-                }
-                // 与 killPane 同款前戏: Esc 给生成中的 CLI 一个收束落盘的机会。
-                await runTmux(["send-keys", "-t", pane, "Escape"]);
-                await sleep(250);
-                const r = await runTmux(["kill-pane", "-t", pane]);
-                if (r.code !== 0) {
-                    skipped[target] = `kill-pane failed: ${r.stdout.slice(-120) || r.code}`;
+                const why = await putDown(pane);
+                if (why) {
+                    skipped[target] = why;
                     continue;
                 }
                 if (a)
@@ -5342,7 +5384,7 @@ export const startMirror = (deps) => {
                 over -= 1;
                 log.info({ target, sessionId, pane, idleHours: Math.round((now - mtime) / 3.6e6), max }, "mirror pane cap — pane killed, binding kept");
             }
-            return { reaped, skipped };
+            return { reaped, orphans, skipped };
         },
         submitPane: async (target) => {
             const a = byTarget.get(target);