aiterm-mcp 0.61.9 → 0.61.10

package/CHANGELOG.md

@@ -7,6 +7,18 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.61.10] - 2026-10-23
+
+### Validation scope
+
+- This release is accepted for the owner-requested Linux image integration with passing Linux/macOS product suites and the SDK/MCP consumer contract. Hosted Windows validation remains unresolved: the full run reported 887 passed, 709 skipped and 8 failed; focused comparisons against the preceding public source also failed. Do not treat this release as a completed three-OS acceptance. No Windows support contract or native effect retry policy is changed.
+
+### Fixed
+
+- Managed native session listings re-observe only an exact `server exited unexpectedly` failure for up to 500 ms. They still require scoped absence evidence or a valid native listing, retain unknown on ambiguous results, and preserve the existing issued slot and generation checks without repeating native effects.
+- `agent_configure` reads a Claude Code seat's busy state with the same harness judgment as `pty_observe` and the ready gate. A stale "esc to interrupt" left in the conversation area no longer keeps a stopped seat waiting as if a turn were running.
+- `pty_list` and `pty_observe` re-read the tmux session listing for up to 500 ms when the server reports `server exited unexpectedly`, which macOS returns right after the last session closes and before the server finishes exiting. If the listing still cannot be read, the error is returned as before.
+
 ## [0.61.9] - 2026-10-23
 
 ### Fixed
@@ -2770,7 +2782,8 @@ prototype (preserved under `prototype/python/` as the porting source and referen
   `ubuntu-latest` for Node 18/20/22, publishing to npm on `v*` tags with
   provenance.
 
-[Unreleased]: https://github.com/kitepon/aiterm-mcp/compare/v0.61.9...HEAD
+[Unreleased]: https://github.com/kitepon/aiterm-mcp/compare/v0.61.10...HEAD
+[0.61.10]: https://github.com/kitepon/aiterm-mcp/compare/v0.61.9...v0.61.10
 [0.61.9]: https://github.com/kitepon/aiterm-mcp/compare/v0.61.8...v0.61.9
 [0.61.8]: https://github.com/kitepon/aiterm-mcp/compare/v0.61.7...v0.61.8
 [0.61.7]: https://github.com/kitepon/aiterm-mcp/compare/v0.61.6...v0.61.7

package/README.ja.md

@@ -209,7 +209,7 @@ runtime-error store は canonical dotagents config の `collection.enabled: true
 場合だけ収集し、既定OFF、network送信は行いません。tag起点CIのnpm provenance（OIDC Trusted
 Publishing）で公開し、GitHub Release が Official MCP Registry を再登録します。
 
-**状態:** 開発継続中 · 現行公開版 **v0.61.9** · 動作対象は Linux · WSL2 · macOS · Windows ネイティブ · MIT · [変更履歴](CHANGELOG.md)。
+**状態:** 開発継続中 · 現行公開版 **v0.61.10** · 動作対象は Linux · WSL2 · macOS · Windows ネイティブ · MIT · [変更履歴](CHANGELOG.md)。
 
 ### 更新と巻き戻し
 
@@ -516,6 +516,7 @@ sessionへ登録する名前の配列で、`pty_list({ env_keys: ["JOB_OWNER"] }
 登録していない値と未設定値はnull。
 一覧の`aiterm.pty-list-result.v1`は`observed_at`と`sessions`を持ち、各行に`session_id`、`current_command`、
 `attached`、`width`、`height`、`harness`、`environment`を返す。従来のtextも維持する。
+最後のsessionを閉じた直後は、終了中のserverの一覧を最長500msで再観測する場合がある。管理されたnative操作は、正常一覧または同socketの厳密な不在を確認できなければunknownを保つ。closeや送信は再実行しない。
 
 `pty_observe({ session_id, cursor? })`は`aiterm.pty-observe-result.v1`で`exists`、`observed_at`、
 `state`（busy／idle／blocked／dead／missing／unknown）、`reason`、`pane_alive`、`harness_alive`を返す。

package/README.md

@@ -227,7 +227,7 @@ collection is off by default and performs no network I/O. It ships via
 tag-triggered CI with npm provenance (OIDC Trusted Publishing); the GitHub
 Release re-registers the Official MCP Registry entry.
 
-**Status:** actively maintained · current public release **v0.61.9** · runs on Linux · WSL2 · macOS · native Windows (tmux on POSIX, the tmux-CLI-compatible [psmux](https://github.com/psmux/psmux) on native Windows — no WSL required) · MIT · see the [CHANGELOG](CHANGELOG.md).
+**Status:** actively maintained · current public release **v0.61.10** · runs on Linux · WSL2 · macOS · native Windows (tmux on POSIX, the tmux-CLI-compatible [psmux](https://github.com/psmux/psmux) on native Windows — no WSL required) · MIT · see the [CHANGELOG](CHANGELOG.md).
 
 ### Update and rollback
 
@@ -551,6 +551,7 @@ On top of that sits a productized layer a raw tmux bridge doesn't have: **token-
 `pty_list({ env_keys: ["JOB_OWNER"] })` returns only registered non-secret values in `environment`; unregistered and missing values are null.
 Its `aiterm.pty-list-result.v1` receipt contains `observed_at` and `sessions`, whose entries include `session_id`,
 `current_command`, `attached`, `width`, `height`, `harness`, and `environment`. Existing text remains available.
+After the last session closes, session listing may briefly re-observe an exiting server for up to 500 ms. Managed native operations retain unknown unless the listing or exact scoped absence can be confirmed; this never repeats a close or send.
 
 `pty_observe({ session_id, cursor? })` returns `aiterm.pty-observe-result.v1` with `exists`, `observed_at`, `state`
 (busy/idle/blocked/dead/missing/unknown), `reason`, `pane_alive`, and `harness_alive`. `pane_process` and `harness_process`

package/dist/agent-screen-observation.js

@@ -0,0 +1,287 @@
+import { AitermError } from "./errors.js";
+import { grokTuiReady, grokTuiBusy, grokPaneObservation, grokLaunchBlockingDialog, assertGrokSandboxNotRejected, GROK_COMPOSER_MARKER_RE } from "./harnesses/grok.js";
+import { codexTuiReady, codexTuiBusy, codexPaneObservation, codexRateLimitModelSwitchDialog, CODEX_COMPOSER_MARKER_RE } from "./harnesses/codex.js";
+import { claudeTuiReady, claudeTuiBusy, claudePaneObservation, claudeLoginMethodMenu, claudeStartupAction, CLAUDE_COMPOSER_MARKER_RE } from "./harnesses/claude.js";
+import { cursorTuiReady, cursorTuiBusy, cursorPaneObservation, CURSOR_COMPOSER_CONTENT_MARKER_RE } from "./harnesses/cursor.js";
+const AGENT_DONE_SETTLE_MIN_MS = 250;
+const AGENT_DONE_SCREEN_SETTLE_POLL_MS = 100;
+const AGENT_DONE_SCREEN_SETTLE_MAX_POLLS = 5;
+const AGENT_DONE_SCREEN_SETTLE_MIN_SAMPLES = 3;
+export const AGENT_TUI_READY_TIMEOUT_MS = 30_000;
+// 画面が起動コマンドの表示のまま（agentがまだ何も描いていない）で上の待ちが切れた時に、起動から待つ上限。
+// Codex親の既定のtool timeout（60秒）の内側に収める。越えると親には時間切れに見えたまま子へ初手が送られる。
+export const AGENT_TUI_FIRST_DRAW_TIMEOUT_MS = 50_000;
+const LAUNCH_ECHO_TAIL_CHARS = 24;
+const LAUNCH_ECHO_MIN_CHARS = 8;
+export const AGENT_TUI_READY_POLL_MS = 500;
+const AGENT_TUI_READY_STABLE_SAMPLES = 11;
+// 上の11回は、起動の途中に一瞬だけ出る入力欄を採らないための確かめ（ADR 0014）。席のharnessが自分の記録で
+// 「この起動でturnを1つ終えていて、次のturnを始めていない」と示している時は起動の途中ではないので、短い間隔で
+// 続けて確かめるだけで送る（ADR 0097）。記録で示せない時は、今までどおり11回。
+const AGENT_TUI_SETTLED_STABLE_SAMPLES = 2;
+const AGENT_TUI_SETTLED_POLL_MS = 100;
+// submit座礁観測（dispatch後にcomposerへ送信textが残存していないかの有界チェック）
+const AGENT_SUBMIT_RESIDUE_DELAY_MS = 250;
+const AGENT_SUBMIT_RESIDUE_POLL_MS = 300;
+const AGENT_SUBMIT_RESIDUE_MAX_SAMPLES = 5;
+const AGENT_SUBMIT_RESIDUE_TAIL_CHARS = 32;
+const AGENT_SUBMIT_RESIDUE_MIN_TAIL_CHARS = 8;
+const CURSOR_PROMPT_VISIBLE_POLL_MS = 100;
+const CURSOR_PROMPT_VISIBLE_MAX_SAMPLES = 50;
+let agentTuiReadyStableSamplesTestOverride = null;
+/**
+ * harness別の画面の読みは、この振り分けだけが持つ。coreの観測・起動準備・送信口は`kind`を渡して呼び、
+ * どのharnessの関数を呼ぶかを自分で選ばない。読みの中身（入力欄・動作中・承認待ちの印）は各harness moduleが所有する。
+ */
+export function agentPaneObservation(kind, screen) {
+    if (kind === "claude")
+        return claudePaneObservation(screen);
+    if (kind === "codex")
+        return codexPaneObservation(screen);
+    if (kind === "cursor")
+        return cursorPaneObservation(screen);
+    return grokPaneObservation(screen);
+}
+export function isAgentTuiReady(kind, screen) {
+    if (kind === "claude")
+        return claudeTuiReady(screen);
+    if (kind === "codex")
+        return codexTuiReady(screen);
+    if (kind === "cursor")
+        return cursorTuiReady(screen);
+    return grokTuiReady(screen);
+}
+// 動作中の印は各harnessの判定を使う（agentPaneObservationがbusyを返す読みと同じ）。以前はClaudeとCursorだけ、ここに
+// 別の正規表現があった。Claudeは会話欄に残った「esc to interrupt」を動作中と読み、止まった席の設定替え（agent_configure）
+// が入力待ちを待ち続けていた。同じ画面を2つの読みで判断しない。
+export function isAgentTuiBusy(kind, screen) {
+    if (kind === "claude")
+        return claudeTuiBusy(screen);
+    if (kind === "codex")
+        return codexTuiBusy(screen);
+    if (kind === "cursor")
+        return cursorTuiBusy(screen);
+    return grokTuiBusy(screen);
+}
+// ready gate 用: 入力欄マーカーがあっても busy 表示中は ready と数えない。
+// frontend 推定（inferAgentFrontend）は「agent TUI が前面か」を見るだけなので isAgentTuiReady のまま。
+export function isAgentTuiIdleReady(kind, screen) {
+    return agentPaneObservation(kind, screen).state === "idle";
+}
+// 起動側が明示応答すべき既知UI。ここで自動承認せず、ready timeoutを待たずに
+// `initial_prompt=not_sent`を返してsessionを生かしたままcallerへ制御を戻す。
+function isAgentTuiActionRequired(kind, screen) {
+    if (kind === "grok")
+        return grokLaunchBlockingDialog(screen) !== null;
+    if (kind === "codex") {
+        return codexPaneObservation(screen).state === "blocked";
+    }
+    if (kind === "claude") {
+        return /new MCP servers? found in this project/iu.test(screen)
+            || claudeLoginMethodMenu(screen)
+            || claudeStartupAction(screen, false)?.kind === "initial_theme_selected"
+            || screen.includes("Is this a project you created or one you trust")
+            || screen.includes("trust this folder")
+            || (screen.includes("Claude Code running in Bypass Permissions mode") && screen.includes("Yes, I accept"));
+    }
+    return false;
+}
+/**
+ * 画面の末尾が、Aitermが打った起動コマンドの表示のままかを見る（純粋関数）。
+ * 混んだ端末ではagentのTUIが描かれる前にready待ちが切れる。画面の最後の文字列が起動コマンドの一部なら、
+ * agentはまだ何も描いていない。空白と行頭の継続prompt（`> `／`>> `）は端末の折返しとshellの表示なので比べない。
+ */
+export function launchEchoIsLastOnScreen(screen, launchLine) {
+    const command = launchLine.replace(/\s+/g, "");
+    const lines = screen.split("\n");
+    let tail = "";
+    for (let i = lines.length - 1; i >= 0 && tail.length < LAUNCH_ECHO_TAIL_CHARS; i--) {
+        tail = lines[i].replace(/^\s*>{1,2}(?:\s|$)/, "").replace(/\s+/g, "") + tail;
+    }
+    tail = tail.slice(-LAUNCH_ECHO_TAIL_CHARS);
+    return tail.length >= LAUNCH_ECHO_MIN_CHARS && command.includes(tail);
+}
+export async function waitAgentTuiReadyImpl(kind, sample, sleepFn, opts = {}) {
+    const timeoutMs = opts.timeoutMs ?? AGENT_TUI_READY_TIMEOUT_MS;
+    const pollMs = opts.pollMs ?? AGENT_TUI_READY_POLL_MS;
+    const stableSamples = opts.stableSamples ?? agentTuiReadyStableSamplesTestOverride ?? AGENT_TUI_READY_STABLE_SAMPLES;
+    const settledSamples = opts.settledSamples ?? AGENT_TUI_SETTLED_STABLE_SAMPLES;
+    const settledPollMs = opts.settledPollMs ?? AGENT_TUI_SETTLED_POLL_MS;
+    const now = opts.now ?? (() => performance.now());
+    const start = now();
+    let deadline = start + timeoutMs;
+    // 呼ぶ側が既定より短い待ちを指定した時は、その待ちを守る（延ばさない）。
+    const firstDrawDeadline = timeoutMs >= AGENT_TUI_READY_TIMEOUT_MS ? start + AGENT_TUI_FIRST_DRAW_TIMEOUT_MS : deadline;
+    let samples = 0;
+    let readyStreak = 0;
+    let settledStreak = 0;
+    let lastScreen = "";
+    for (;;) {
+        lastScreen = sample();
+        samples++;
+        if (kind === "grok")
+            assertGrokSandboxNotRejected(lastScreen);
+        if (isAgentTuiIdleReady(kind, lastScreen)) {
+            readyStreak++;
+            if (readyStreak >= stableSamples)
+                return { ready: true, samples, lastScreen };
+            if (opts.turnSettled?.()) {
+                settledStreak++;
+                if (settledStreak >= settledSamples)
+                    return { ready: true, samples, lastScreen };
+            }
+            else if (settledStreak > 0) {
+                // 記録の上でturnが始まった。短い間隔で数えた分を、11回の確かめへ持ち越さない。
+                settledStreak = 0;
+                readyStreak = 1;
+            }
+        }
+        else {
+            readyStreak = 0;
+            settledStreak = 0;
+            // The one keep-current key can be accepted before the TUI redraws.
+            // Wait for that exact handled modal to disappear; never resend its key.
+            const recoveringModal = opts.codexRateLimitRecovery && kind === "codex"
+                && codexRateLimitModelSwitchDialog(lastScreen) !== null;
+            if (isAgentTuiActionRequired(kind, lastScreen) && !recoveringModal)
+                return { ready: false, samples, lastScreen };
+        }
+        if (now() >= deadline) {
+            // TUIが一度も描かれていない間に切れた時だけ、決めた上限まで待ちを延ばす。描かれた後の画面は今までどおり扱う。
+            if (deadline >= firstDrawDeadline || opts.launchLine === undefined
+                || !launchEchoIsLastOnScreen(lastScreen, opts.launchLine))
+                return { ready: false, samples, lastScreen };
+            deadline = firstDrawDeadline;
+        }
+        await sleepFn(settledStreak > 0 ? settledPollMs : pollMs);
+    }
+}
+function normalizeResidueText(s) {
+    return s.replace(/\s+/g, "");
+}
+function agentSubmitResidueTail(text) {
+    // 行単位でなく text 全体の正規化末尾から取る: 最終行が短い prompt（「以上」等の締め行）でも
+    // 直前行の内容を含む末尾 32 codepoint で観測できる。composer は末尾（カーソル位置）を表示し、
+    // submit 済みの transcript echo は長文では先頭側を表示するため、末尾一致は座礁側に偏る。
+    const cps = [...normalizeResidueText(text)];
+    if (cps.length < AGENT_SUBMIT_RESIDUE_MIN_TAIL_CHARS)
+        return null;
+    return cps.slice(-AGENT_SUBMIT_RESIDUE_TAIL_CHARS).join("");
+}
+function agentSubmitResidueOnScreen(kind, screen, tail) {
+    // Cursorは送信成立後のactive turnで明示的な停止UIを出す。scrollbackに残った旧「>」を
+    // composerと誤認すると、実際には走っているpromptをsubmit座礁として報告してしまう。
+    if (kind === "cursor" && isAgentTuiBusy(kind, screen))
+        return false;
+    const lines = screen.split("\n");
+    // 入力欄マーカーは ready 判定と同じ記号を行頭基準で探す。submit 済みの transcript echo は
+    // マーカー行より上に出るため、最後のマーカー行以降だけを composer 領域として見る。
+    // grok は Windows native 描画（`>`・実測 1.0.4）も ready 判定と同様に受ける。
+    const markerRe = kind === "codex"
+        ? CODEX_COMPOSER_MARKER_RE
+        : kind === "claude"
+            ? CLAUDE_COMPOSER_MARKER_RE
+            : kind === "cursor"
+                ? CURSOR_COMPOSER_CONTENT_MARKER_RE
+                : GROK_COMPOSER_MARKER_RE;
+    let markerIdx = -1;
+    for (let i = lines.length - 1; i >= 0; i--) {
+        if (markerRe.test(lines[i])) {
+            markerIdx = i;
+            break;
+        }
+    }
+    if (markerIdx < 0)
+        return null;
+    if (kind === "cursor") {
+        // Cursorは長いpasteを入力欄の先頭側だけ表示し、矢印の次行へ本文を折り返すことがある。
+        // 最新composerから下部statusを除いた領域にplaceholder以外の本文が残っていれば、
+        // 送信文の末尾が画面外でも未submitの陽性証拠とする。
+        const composerLines = lines.slice(markerIdx);
+        composerLines[0] = composerLines[0].replace(/^\s*(?:->|>|→)\s*/u, "");
+        while (composerLines.at(-1)?.trim() === "")
+            composerLines.pop();
+        if (/^\s*\/\S.*\s·\s\S/u.test(composerLines.at(-1) ?? ""))
+            composerLines.pop();
+        while (composerLines.at(-1)?.trim() === "")
+            composerLines.pop();
+        if (/^\s*(?:Auto|Ask)(?:\s|·|$)/iu.test(composerLines.at(-1) ?? ""))
+            composerLines.pop();
+        while (composerLines.at(-1)?.trim() === "")
+            composerLines.pop();
+        const composerText = composerLines.join("\n").trim();
+        if (composerText.length > 0
+            && !/^(?:Add a follow-up|Plan,\s*search,\s*build anything)\b/iu.test(composerText))
+            return true;
+    }
+    return normalizeResidueText(lines.slice(markerIdx).join("")).includes(tail);
+}
+export async function detectAgentSubmitResidueImpl(kind, text, sample, sleepFn, opts = {}) {
+    const tail = agentSubmitResidueTail(text);
+    if (!tail)
+        return { residue: null, samples: 0 };
+    const delayMs = opts.delayMs ?? AGENT_SUBMIT_RESIDUE_DELAY_MS;
+    const pollMs = opts.pollMs ?? AGENT_SUBMIT_RESIDUE_POLL_MS;
+    const maxSamples = opts.maxSamples ?? AGENT_SUBMIT_RESIDUE_MAX_SAMPLES;
+    if (delayMs > 0)
+        await sleepFn(delayMs);
+    let samples = 0;
+    let last = null;
+    for (let i = 0; i < maxSamples; i++) {
+        last = agentSubmitResidueOnScreen(kind, sample(), tail);
+        samples++;
+        // 残存が消えた（または判定不能になった）時点で確定。true だけは描画遅延と区別するため
+        // 全サンプル持続した場合にのみ報告する。
+        if (last !== true)
+            return { residue: last, samples };
+        if (i < maxSamples - 1)
+            await sleepFn(pollMs);
+    }
+    return { residue: true, samples };
+}
+export async function waitCursorPromptVisibleImpl(text, sample, sleepFn, opts = {}) {
+    const tail = agentSubmitResidueTail(text) ?? normalizeResidueText(text);
+    const pollMs = opts.pollMs ?? CURSOR_PROMPT_VISIBLE_POLL_MS;
+    const maxSamples = opts.maxSamples ?? CURSOR_PROMPT_VISIBLE_MAX_SAMPLES;
+    for (let i = 0; i < maxSamples; i++) {
+        if (agentSubmitResidueOnScreen("cursor", sample(), tail) === true) {
+            return { visible: true, samples: i + 1 };
+        }
+        if (i < maxSamples - 1)
+            await sleepFn(pollMs);
+    }
+    return { visible: false, samples: maxSamples };
+}
+export async function settleAgentDoneScreenImpl(sample, sleepFn, opts = {}) {
+    const minDelayMs = opts.minDelayMs ?? AGENT_DONE_SETTLE_MIN_MS;
+    const pollMs = opts.pollMs ?? AGENT_DONE_SCREEN_SETTLE_POLL_MS;
+    const maxPolls = opts.maxPolls ?? AGENT_DONE_SCREEN_SETTLE_MAX_POLLS;
+    const minSamples = opts.minSamples ?? AGENT_DONE_SCREEN_SETTLE_MIN_SAMPLES;
+    if (minDelayMs > 0)
+        await sleepFn(minDelayMs);
+    let prev = sample();
+    let samples = 1;
+    let stableStreak = 1;
+    for (let i = 0; i < maxPolls; i++) {
+        if (pollMs > 0)
+            await sleepFn(pollMs);
+        const current = sample();
+        samples++;
+        if (current.screen === prev.screen && current.logSize === prev.logSize) {
+            stableStreak++;
+            if (samples >= minSamples && stableStreak >= 2)
+                return { unstable: false, samples };
+        }
+        else {
+            stableStreak = 1;
+        }
+        prev = current;
+    }
+    return { unstable: true, samples };
+}
+export function __testSetAgentTuiReadyStableSamples(value) {
+    if (value !== null && (!Number.isInteger(value) || value < 1 || value > 1000)) {
+        throw new AitermError("test ready stable samplesが不正です", 2);
+    }
+    agentTuiReadyStableSamplesTestOverride = value;
+}