kimi-plugin-cc 2.4.1 → 2.4.3

package/.claude-plugin/marketplace.json

@@ -5,7 +5,7 @@
   },
   "metadata": {
     "description": "Kimi for Claude Code: code questions, reviews, fixes, parallel tasks, and MCP tasks through the local Kimi Code CLI.",
-    "version": "2.4.1"
+    "version": "2.4.3"
   },
   "plugins": [
     {

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "kimi",
   "description": "Kimi for Claude Code: code questions, independent reviews, bounded fixes, autonomous goals, parallel tasks, and MCP tasks through the local Kimi Code CLI, with PreToolUse safety hooks.",
-  "version": "2.4.1",
+  "version": "2.4.3",
   "homepage": "https://github.com/linxule/kimi-plugin-cc",
   "repository": "https://github.com/linxule/kimi-plugin-cc",
   "license": "Apache-2.0",

package/CHANGELOG.md

@@ -2,6 +2,21 @@
 
 > **Post-1.0 release history (v1.0.1 -> present) lives in [ROADMAP-TO-GA.md § Post-GA audit log](./ROADMAP-TO-GA.md#post-ga-audit-log)** and the "Version" / "Upstream compat" lines of [AGENTS.md](./AGENTS.md). Docs-only kimi-code compat checkups that don't bump the plugin version (e.g. the 0.14.2 / 0.14.3 patches) are recorded there, not here. Notable releases are summarized below; the GA entry and full pre-GA detail follow.
 
+## 2.4.3 — 2026-10-27
+
+- Certify exact kimi-code 2.3.3 for all eight native-v2 operations after exact-source review, plan and swarm controls, and candidate/production smoke. See [evidence](docs/upstream-2.3.3-certification.md).
+- Add only the reviewed 2.3.3 activation and hook-schema entries. Preserve no-plan enforcement, workspace confinement, concurrency, bounded cleanup and durable recovery; legacy-v1 remains capped at 0.41.x.
+- Expand the compatibility audit to cover the web activation server explicitly, and refresh current compatibility/status documentation without changing installed hosts.
+
+## 2.4.2 — 2026-10-26
+
+- Integrate community PR #22's structured spawn-error classification, preserving contributor history. Bound the additional diagnostic message. Known argument/resource exhaustion no longer receives reinstall advice.
+- Refuse oversized Windows command lines before spawn and add a credential-free compiled-runtime Windows CI lane. This improves diagnosis, not prompt transport; issue #14 remains open for native Windows/community verification and large-review support.
+- Add durable, preview-first `recover-swarm` for exact 2.3.2 interrupted runs, with verified process ownership, pre-mutation session identity, recovery locking, explicit uncertainty refusals and patch preservation. Stop age-based deletion of abandoned worktrees. Exact CLI certification remains unchanged through 2.3.2.
+- Shorten the agent contract sheet and enforce its 32 KiB limit during `check`.
+
+Thanks to **@Nitjsefnie** for PR #22.
+
 ## 2.4.1 — 2026-10-25
 
 - Certify exact kimi-code 2.3.2 for all eight operations using the 15-test/92-assertion source scan, plan-ON/OFF controls, and the final 19-test/125-assertion production smoke. Keep the no-plan construction and legacy-v1 boundary at 0.41.x. See [evidence](docs/upstream-2.3.2-certification.md).

package/README.md

@@ -40,9 +40,9 @@ You need:
 - macOS or Linux
 - Claude Code or Codex with plugin support
 - Node.js 22.5 or newer
-- Kimi Code CLI 2.3.2 (certified; 2.3.1 and 2.3.0 are also certified)
+- Kimi Code CLI 2.3.3 (certified; 2.3.2, 2.3.1 and 2.3.0 are also certified)
 
-Windows is not currently supported. Check your versions in a terminal:
+Windows is not currently supported. Contributors can help with the [no-model Windows diagnostics checks](./docs/windows-validation.md). Check your versions in a terminal:
 
 ```sh
 node --version
@@ -105,10 +105,10 @@ Codex. Published releases are also available as
 both compiled plugin distributions. No Bun, TypeScript, or install scripts are
 needed to use the package. Kimi Code CLI and Node are still prerequisites.
 
-For an explicitly versioned local Claude Code plugin, this example pins the published 2.4.0 package; 2.4.1 is not yet published:
+For an explicitly versioned local Claude Code plugin, this example pins the 2.4.3 package:
 
 ```sh
-npm install --prefix ~/.local/share/kimi-plugin-cc kimi-plugin-cc@2.4.0
+npm install --prefix ~/.local/share/kimi-plugin-cc kimi-plugin-cc@2.4.3
 claude --plugin-dir ~/.local/share/kimi-plugin-cc/node_modules/kimi-plugin-cc
 ```
 
@@ -174,7 +174,7 @@ Use these commands to follow a task:
 | Read the result | `/kimi:result` | `$kimi-result` |
 | Cancel work | `/kimi:cancel` | `$kimi-cancel` |
 
-Without a job ID, cancel stops the latest running task for this repository. If several tasks are running, check which one you want to stop first. Use cancel rather than interrupting the app so write-swarm can save its patch.
+Without a job ID, cancel stops the latest running task for this repository. If several tasks are running, check which one you want to stop first. Use cancel rather than interrupting the app so write-swarm can save its patch. For interrupted runs created by the unreleased recovery implementation, see [preview-first swarm recovery](./docs/swarm-recovery.md); it never applies a patch or deletes abandoned work.
 
 Canceling a task does not undo changes it has already made.
 
@@ -228,7 +228,7 @@ All refusal codes and their remedies are listed in the [refusal-code catalog](./
 
 The tool rules for ask, review, challenge, and read-only swarm deny file writes and shell commands. Rescue and pursue can write files in your workspace. Swarm with `--write` works in a temporary Git worktree. It starts from your latest commit, so it does not include uncommitted changes. It returns a patch without applying it to your project.
 
-For swarm on CLI 2.3.2, the plugin briefly starts an authenticated loopback `kimi web` process to activate a session and another to deactivate swarm mode afterward; one ordinary `kimi -p -r` process still performs the task. Activation counts against the time limit, the access token stays in memory and out of logs, and cleanup failure prevents success. Older certified versions use `kimi -p` directly.
+For swarm on CLI 2.3.2 and 2.3.3, the plugin briefly starts an authenticated loopback `kimi web` process to activate a session and another to deactivate swarm mode afterward; one ordinary `kimi -p -r` process still performs the task. Activation counts against the time limit, the access token stays in memory and out of logs, and cleanup failure prevents success. Older certified versions use `kimi -p` directly.
 
 The plugin does not commit changes. Review changes before you keep or apply them.
 
@@ -264,7 +264,7 @@ Contributors need Bun to run `bun run check`. The detailed engineering documents
 - macOS 或 Linux
 - 支持插件的 Claude Code 或 Codex
 - Node.js 22.5 或更高版本
-- Kimi Code CLI 2.3.2（已认证；2.3.1 和 2.3.0 也已认证）
+- Kimi Code CLI 2.3.3（已认证；2.3.2、2.3.1 和 2.3.0 也已认证）
 
 目前不支持 Windows。在终端检查版本：
 
@@ -427,7 +427,7 @@ codex plugin add kimi@kimi-marketplace
 
 Ask、review、challenge 和只读 swarm 的工具规则禁止写入文件和执行 shell 命令。Rescue 和 pursue 可以修改工作区内的文件。带 `--write` 的 swarm 在临时 Git 工作树中操作。它从最新提交开始，不包含尚未提交的修改。任务结束后返回补丁，不会自动应用到项目。
 
-在 CLI 2.3.2 上运行 swarm 时，插件会短暂启动需要认证的本地回环 `kimi web` 进程来激活会话，并在结束后再次启动以停用 swarm 模式；任务本身仍由一个普通的 `kimi -p -r` 进程执行。激活计入任务时限，访问令牌只保存在内存且不会写入日志；无法确认清理完成时，任务不会报告成功。较早的已认证版本仍直接通过 `kimi -p` 运行。
+在 CLI 2.3.2 和 2.3.3 上运行 swarm 时，插件会短暂启动需要认证的本地回环 `kimi web` 进程来激活会话，并在结束后再次启动以停用 swarm 模式；任务本身仍由一个普通的 `kimi -p -r` 进程执行。激活计入任务时限，访问令牌只保存在内存且不会写入日志；无法确认清理完成时，任务不会报告成功。较早的已认证版本仍直接通过 `kimi -p` 运行。
 
 插件不会提交修改。保留或应用修改前，请先审查。
 
@@ -463,7 +463,7 @@ Il vous faut :
 - macOS ou Linux
 - une version de Claude Code ou de Codex prenant en charge les plugins
 - Node.js 22.5 ou une version ultérieure
-- Kimi Code CLI 2.3.2 (certifiée ; les versions 2.3.1 et 2.3.0 sont également certifiées)
+- Kimi Code CLI 2.3.3 (certifiée ; les versions 2.3.2, 2.3.1 et 2.3.0 sont également certifiées)
 
 Windows n'est pas pris en charge actuellement. Vérifiez les versions dans un terminal :
 
@@ -626,7 +626,7 @@ Tous les codes de refus et leurs remèdes sont répertoriés dans le [catalogue
 
 Les règles des modes ask, review, challenge et swarm en lecture seule interdisent les écritures et les commandes shell. Rescue et pursue peuvent modifier les fichiers de votre espace de travail. Swarm avec `--write` travaille dans un worktree Git temporaire. Il part du dernier commit et n'inclut donc pas les modifications non commitées. Il renvoie un patch sans l'appliquer à votre projet.
 
-Avec le CLI 2.3.2, les tâches swarm lancent brièvement un processus `kimi web` authentifié sur l'interface de boucle locale pour activer la session, puis un second pour désactiver le mode swarm ; un seul processus `kimi -p -r` ordinaire exécute toujours la tâche. L'activation est comptée dans la limite de temps, le jeton reste en mémoire et hors des journaux, et l'échec du nettoyage empêche de déclarer la réussite. Les versions certifiées antérieures utilisent directement `kimi -p`.
+Avec le CLI 2.3.2 et 2.3.3, les tâches swarm lancent brièvement un processus `kimi web` authentifié sur l'interface de boucle locale pour activer la session, puis un second pour désactiver le mode swarm ; un seul processus `kimi -p -r` ordinaire exécute toujours la tâche. L'activation est comptée dans la limite de temps, le jeton reste en mémoire et hors des journaux, et l'échec du nettoyage empêche de déclarer la réussite. Les versions certifiées antérieures utilisent directement `kimi -p`.
 
 Le plugin ne crée aucun commit. Examinez les modifications avant de les conserver ou de les appliquer.
 
@@ -662,7 +662,7 @@ Les contributeurs ont besoin de Bun pour exécuter `bun run check`. La documenta
 - macOS または Linux
 - プラグインに対応した Claude Code または Codex
 - Node.js 22.5 以降
-- Kimi Code CLI 2.3.2（認証済み。2.3.1 と 2.3.0 も認証済み）
+- Kimi Code CLI 2.3.3（認証済み。2.3.2、2.3.1 と 2.3.0 も認証済み）
 
 現在、Windows には対応していません。ターミナルでバージョンを確認します。
 
@@ -825,7 +825,7 @@ codex plugin add kimi@kimi-marketplace
 
 Ask、review、challenge、読み取り専用 swarm のツールルールでは、ファイルへの書き込みとシェルコマンドを禁止しています。Rescue と pursue はワークスペース内のファイルを変更できます。`--write` を指定した swarm は、一時的な Git ワークツリーで作業します。最新のコミットから開始するため、未コミットの変更は含まれません。パッチを返すだけで、プロジェクトには自動適用しません。
 
-CLI 2.3.2 の swarm では、セッションを有効化するため、そして終了後に swarm モードを無効化するため、認証付きのローカルループバック `kimi web` プロセスをそれぞれ短時間だけ起動します。タスク自体は従来どおり 1 つの `kimi -p -r` プロセスが実行します。有効化も制限時間に含まれ、トークンはメモリー内にのみ保持されてログには出ません。後処理を確認できない場合、タスクは成功と報告されません。以前の認証済みバージョンでは、引き続き `kimi -p` を直接使います。
+CLI 2.3.2 / 2.3.3 の swarm では、セッションを有効化するため、そして終了後に swarm モードを無効化するため、認証付きのローカルループバック `kimi web` プロセスをそれぞれ短時間だけ起動します。タスク自体は従来どおり 1 つの `kimi -p -r` プロセスが実行します。有効化も制限時間に含まれ、トークンはメモリー内にのみ保持されてログには出ません。後処理を確認できない場合、タスクは成功と報告されません。以前の認証済みバージョンでは、引き続き `kimi -p` を直接使います。
 
 プラグインは変更をコミットしません。変更を残す、または適用する前に内容を確認してください。
 

package/dist/cli-client.js

@@ -7,6 +7,7 @@
 import { spawn } from "node:child_process";
 import { appendFile, mkdir } from "node:fs/promises";
 import path from "node:path";
+import { assertPromptFitsCommandLine } from "./prompt-size.js";
 import { createHookEscapeMonitor } from "./hook-escape.js";
 import { RuntimeError, formatError } from "./errors.js";
 import { assertNoUnsafeExperimentalSelector, assertExecutionPlanMatchesSpawn, } from "./kimi-engine.js";
@@ -29,6 +30,14 @@ const DESCENDANT_SNAPSHOT_TIMEOUT_MS = 1_000;
 /** Hard bound for each batched identity revalidation pass. */
 const IDENTITY_REVALIDATION_TIMEOUT_MS = 250;
 const DESCENDANT_SNAPSHOT_TIMEOUT = Symbol("descendant-snapshot-timeout");
+function spawnErrorDetails(error) {
+    const cause = error;
+    return {
+        spawn_code: cause?.code,
+        spawn_errno: cause?.errno,
+        spawn_message: formatError(error),
+    };
+}
 export async function runCliPrompt(opts) {
     // Pre-aborted signal: refuse to spawn at all.
     if (opts.signal?.aborted === true) {
@@ -60,6 +69,7 @@ export async function runCliPrompt(opts) {
         });
     }
     const args = buildArgs(opts);
+    assertPromptFitsCommandLine(opts.command, args);
     const env = buildKimiCliEnvironment(opts);
     if (opts.logPath !== undefined) {
         await mkdir(path.dirname(opts.logPath), { recursive: true }).catch(() => {
@@ -82,8 +92,8 @@ export async function runCliPrompt(opts) {
     }
     catch (err) {
         throw new RuntimeError("CLI_SPAWN_FAILED", `Failed to spawn kimi: ${formatError(err)}`, "cli-client.spawn", err instanceof Error
-            ? { cause: err, details: { command: opts.command, args } }
-            : { details: { command: opts.command, args } });
+            ? { cause: err, details: { command: opts.command, args, ...spawnErrorDetails(err) } }
+            : { details: { command: opts.command, args, ...spawnErrorDetails(err) } });
     }
     // Pre-attach an 'error' listener immediately. Bun (and Node) can emit
     // 'error' asynchronously between spawn returning and our Promise body
@@ -550,8 +560,8 @@ export async function runCliPrompt(opts) {
         child.on("error", (err) => {
             appendLogLine({ event: "process_error", message: formatError(err) });
             settle("reject", new RuntimeError("CLI_PROCESS_ERROR", `kimi subprocess error: ${formatError(err)}`, "cli-client.process", err instanceof Error
-                ? { cause: err, details: errorDetails() }
-                : { details: errorDetails() }));
+                ? { cause: err, details: { ...errorDetails(), ...spawnErrorDetails(err) } }
+                : { details: { ...errorDetails(), ...spawnErrorDetails(err) } }));
         });
         if (earlyErrors.length > 0) {
             const captured = earlyErrors[0];
@@ -562,7 +572,7 @@ export async function runCliPrompt(opts) {
                     early: true,
                     additional_count: earlyErrors.length - 1,
                 });
-                settle("reject", new RuntimeError("CLI_PROCESS_ERROR", `kimi subprocess error: ${formatError(captured)}`, "cli-client.process", { cause: captured, details: errorDetails() }));
+                settle("reject", new RuntimeError("CLI_PROCESS_ERROR", `kimi subprocess error: ${formatError(captured)}`, "cli-client.process", { cause: captured, details: { ...errorDetails(), ...spawnErrorDetails(captured) } }));
             });
         }
         child.on("close", async (exitCode, signal) => {
@@ -643,8 +653,8 @@ export async function runCliPrompt(opts) {
         }
         catch (err) {
             settle("reject", new RuntimeError("CLI_PROCESS_ERROR", `kimi subprocess stream setup failed: ${formatError(err)}`, "cli-client.streams", err instanceof Error
-                ? { cause: err, details: errorDetails() }
-                : { details: errorDetails() }));
+                ? { cause: err, details: { ...errorDetails(), ...spawnErrorDetails(err) } }
+                : { details: { ...errorDetails(), ...spawnErrorDetails(err) } }));
         }
     });
 }

package/dist/commands/cancel.js

@@ -36,6 +36,14 @@ export async function runCancel(argv, context) {
             (context.mcpCompanionIdentity === undefined || context.mcpKimiIdentity === undefined)) {
             throw new RuntimeError("MCP_CANCEL_REQUIRES_BINDING", `MCP job ${job.job_id} requires MCP job-control cancellation with recorded identities; use tasks/cancel or kimi_job_cancel.`, "mcp.cancel");
         }
+        const recovery = store.getSwarmRecovery(job.job_id)?.record;
+        if (recovery) {
+            const identity = recovery.owner;
+            const verified = identity && identity.pid === job.pid ? await revalidateProcessIdentities([identity]) : undefined;
+            if (!verified?.complete || verified.groupChanged?.length || verified.identities.length !== 1) {
+                throw new RuntimeError("SWARM_RECOVERY_REQUIRED", `Recorded swarm ownership is no longer live or verifiable. Preview companion.sh recover-swarm ${job.job_id}; no process was signalled.`, "cancel.identity");
+            }
+        }
         if (job.pid === null && job.kimi_pid === null) {
             throw new RuntimeError("CANCEL_NOT_SUPPORTED", `Job ${job.job_id} does not have a recorded process id to cancel.`, "cancel.runtime");
         }
@@ -75,7 +83,10 @@ export async function runCancel(argv, context) {
         }
         await assertMcpCompanionIdentity(job, context.mcpCompanionIdentity);
         await assertMcpKimiIdentity(job, context.mcpKimiIdentity);
-        signalJobProcesses(job, "SIGTERM");
+        if (recovery)
+            await assertMcpCompanionIdentity(job, recovery.owner);
+        // The live swarm owner owns child teardown, including web phases after print exit.
+        signalJobProcesses(job, "SIGTERM", { companion: true, kimi: !recovery });
         const cancelled = await waitForCancellation(paths, job.job_id);
         if (cancelled) {
             return `${JSON.stringify({
@@ -84,10 +95,10 @@ export async function runCancel(argv, context) {
                 message: "Cancellation completed.",
             }, null, 2)}\n`;
         }
-        const companionMatches = await assertMcpCompanionIdentity(job, context.mcpCompanionIdentity, true);
+        const companionMatches = await assertMcpCompanionIdentity(job, recovery ? recovery.owner : context.mcpCompanionIdentity, true);
         const kimiMatches = await assertMcpKimiIdentity(job, context.mcpKimiIdentity, true);
         if (companionMatches || kimiMatches) {
-            signalJobProcesses(job, "SIGKILL", { companion: companionMatches, kimi: kimiMatches });
+            signalJobProcesses(job, "SIGKILL", { companion: companionMatches, kimi: !recovery && kimiMatches });
         }
         await withJobStore(paths, async (forcedStore) => {
             const current = forcedStore.getJob(job.job_id);

package/dist/commands/recover-swarm.js

@@ -0,0 +1,135 @@
+import { RECOVER_SWARM_SUPPORTED_FLAGS } from "../parsing.js";
+import { randomUUID } from "node:crypto";
+import { lstat, realpath, writeFile } from "node:fs/promises";
+import path from "node:path";
+import { createCliCancellationHandlers } from "../cli-cancellation.js";
+import { RuntimeError } from "../errors.js";
+import { captureWorktreePatch, resolveRepoIdentity } from "../git.js";
+import { verifyHookInstalled } from "../hooks/install.js";
+import { markJobFailed } from "../jobs.js";
+import { JobStore } from "../job-store.js";
+import { executionPlanFromPersisted, prepareKimiExecutionPlan } from "../kimi-engine.js";
+import { resolveKimiHome } from "../kimi-home.js";
+import { ensurePluginPaths, resolvePluginPaths } from "../paths.js";
+import { probeProcessIdentity } from "../process-tree.js";
+import { deactivateSwarmSession } from "../swarm-activation.js";
+import { assertRecoveryOwnerGone, finishRecoveryRecord, recordRecoveryEvent, recoverLaunch, recoveryError } from "../swarm-recovery.js";
+export async function runRecoverSwarm(argv, context) {
+    const positional = argv.filter((arg) => !arg.startsWith("-"));
+    if (positional.length !== 1 || argv.some((arg) => arg.startsWith("-") && arg !== RECOVER_SWARM_SUPPORTED_FLAGS) || argv.filter((arg) => arg === RECOVER_SWARM_SUPPORTED_FLAGS).length > 1) {
+        throw new RuntimeError("INVALID_ARGS", `Usage: companion.sh recover-swarm <job-id> [${RECOVER_SWARM_SUPPORTED_FLAGS}]. Defaults to preview.`, "swarm.recovery.args");
+    }
+    const jobId = positional[0];
+    const apply = argv.includes(RECOVER_SWARM_SUPPORTED_FLAGS);
+    const paths = resolvePluginPaths(context.env);
+    await ensurePluginPaths(paths);
+    const repo = await resolveRepoIdentity(context.cwd);
+    const store = new JobStore(paths);
+    let claim;
+    const handlers = createCliCancellationHandlers();
+    const signal = AbortSignal.any([handlers.signal, AbortSignal.timeout(30_000)]);
+    try {
+        const job = store.getJob(jobId);
+        const saved = store.getSwarmRecovery(jobId);
+        if (!job || job.repo_id !== repo.repoId || !saved || saved.record.schema !== 1 || saved.record.jobId !== jobId) {
+            throw recoveryError("No recovery record owned by this repository. Older runs require manual inspection.");
+        }
+        let record = saved.record;
+        const summary = () => ({ job_id: jobId, mode: apply ? "apply" : "preview", recovery_complete: record.complete,
+            session_id: record.session?.sessionId ?? null, session_cleanup_pending: record.sessionDirty,
+            launches: record.launches, worktree: record.workspaceRoot, patch_path: record.patchPath,
+            next_step: record.complete ? "No recovery is needed. Retained worktrees may be inspected and removed manually."
+                : `Inspect this record; use companion.sh recover-swarm ${jobId} --apply only after the original job has stopped.` });
+        const settleAbandonedJob = async () => {
+            if (job.status === "running")
+                await markJobFailed(store, paths, job, new RuntimeError("SWARM_OWNER_DISAPPEARED", "The original swarm companion stopped without reporting a result; resource recovery completed, not the task.", "swarm.recovery"), "Abandoned swarm resources recovered; task did not complete.", { phase: "failed" });
+        };
+        if (!apply)
+            return `${JSON.stringify(summary(), null, 2)}\n`;
+        if (record.complete) {
+            if (job.status === "running") {
+                await assertRecoveryOwnerGone(record);
+                await settleAbandonedJob();
+            }
+            return `${JSON.stringify(summary(), null, 2)}\n`;
+        }
+        await assertRecoveryOwnerGone(record);
+        if (saved.owner !== null) {
+            const owner = JSON.parse(saved.owner);
+            const live = await probeProcessIdentity(owner.pid);
+            if (!live.complete || live.identity?.startToken === owner.startToken)
+                throw recoveryError("Another recovery owner is alive or cannot be verified.");
+        }
+        const self = await probeProcessIdentity(process.pid);
+        if (!self.complete || !self.identity)
+            throw recoveryError("Cannot establish recovery ownership on this host.");
+        claim = JSON.stringify(self.identity);
+        if (!store.claimSwarmRecovery(jobId, saved.owner, claim)) {
+            claim = undefined;
+            throw recoveryError("Another recovery acquired this job.");
+        }
+        // Another owner may have completed and released the claim between our first read
+        // and the CAS (null -> owner -> null). Never overwrite that newer journal.
+        record = store.getSwarmRecovery(jobId).record;
+        if (record.complete) {
+            await settleAbandonedJob();
+            return `${JSON.stringify(summary(), null, 2)}\n`;
+        }
+        const expectedRoot = job.operation_kind === "swarm-write"
+            ? path.join(paths.worktreesDir, job.repo_id, `swarm-write-${job.job_id}`) : null;
+        if (record.workspaceRoot !== expectedRoot || record.cwd !== (expectedRoot ?? job.cwd)
+            || record.kimiHome !== resolveKimiHome(context.env, record.cwd)) {
+            throw recoveryError("Recovery workspace or Kimi home differs from the recorded job. Use its original home and repository; do not guess a session.");
+        }
+        if (expectedRoot) {
+            const st = await lstat(expectedRoot);
+            if (!st.isDirectory() || st.isSymbolicLink())
+                throw recoveryError("Recovery worktree must be the original non-symlink directory.");
+            const parent = await realpath(path.dirname(expectedRoot));
+            if (await realpath(expectedRoot) !== path.join(parent, path.basename(expectedRoot)))
+                throw recoveryError("Recovery worktree path is redirected.");
+        }
+        const persist = () => store.saveSwarmRecovery(jobId, record);
+        for (const launch of record.launches)
+            await recoverLaunch(launch, persist, signal);
+        if (record.sessionDirty) {
+            if (!record.session || !/^[\w-]+$/.test(record.session.sessionId) || !/^[\w-]+$/.test(record.session.workspaceId))
+                throw recoveryError("Recovery session identity is missing or invalid.");
+            const installed = await verifyHookInstalled(context.env);
+            if (!installed.installed)
+                throw recoveryError(`Managed hook verification failed: ${installed.reason ?? "unknown"}. Recovery does not install or relax hooks.`);
+            const plan = executionPlanFromPersisted(job);
+            const current = await prepareKimiExecutionPlan({ operationKind: plan.operationKind, cwd: record.cwd, env: context.env });
+            if (current.command !== plan.command || current.kimiVersion !== plan.kimiVersion || JSON.stringify(current.prefixArgs) !== JSON.stringify(plan.prefixArgs))
+                throw recoveryError("The exact recorded CLI is not selected. Restore that certified binary before session cleanup.");
+            signal.throwIfAborted();
+            await deactivateSwarmSession({ cwd: record.cwd, env: context.env, command: current.command,
+                prefixArgs: current.prefixArgs, executionPlan: current, commandLabel: expectedRoot ? "swarm-write" : "swarm",
+                trustedWorkspaceRoot: expectedRoot ?? undefined, ...record.session, signal,
+                onRecoveryEvent: (event) => recordRecoveryEvent(store, record, event) });
+        }
+        signal.throwIfAborted();
+        if (expectedRoot) {
+            // Capture without applying or deleting. A crash after the write leaves a recoverable artifact.
+            const patch = await captureWorktreePatch(expectedRoot, signal);
+            const patchPath = path.join(paths.artifactsDir, `swarm-recovered-${jobId}-${randomUUID()}.patch`);
+            await writeFile(patchPath, patch, { flag: "wx", mode: 0o600 });
+            record.patchPath = patchPath;
+        }
+        finishRecoveryRecord(store, record);
+        if (!record.complete)
+            throw recoveryError("Recovery remains incomplete; evidence and worktree were retained.");
+        await settleAbandonedJob();
+        return `${JSON.stringify(summary(), null, 2)}\n`;
+    }
+    finally {
+        handlers.dispose();
+        try {
+            if (claim !== undefined)
+                store.claimSwarmRecovery(jobId, claim, null);
+        }
+        finally {
+            store.close();
+        }
+    }
+}

package/dist/commands/status.js

@@ -22,6 +22,6 @@ export async function runStatus(argv, context) {
             throw new RuntimeError("JOB_NOT_FOUND", "No matching job was found for status.", "status.lookup");
         }
         const job = await reconcileHistoricalJobProvenance(store, found);
-        return `${JSON.stringify({ ...job, thinking_effort_requested: job.thinking_effort ?? null, hook_escapes: parseHookEscapes(job.hook_escapes), budget_source: renderBudgetSource(job) }, null, 2)}\n`;
+        return `${JSON.stringify({ ...job, swarm_recovery: store.getSwarmRecovery(job.job_id)?.record ?? null, thinking_effort_requested: job.thinking_effort ?? null, hook_escapes: parseHookEscapes(job.hook_escapes), budget_source: renderBudgetSource(job) }, null, 2)}\n`;
     });
 }

package/dist/commands/swarm.js

@@ -1,19 +1,22 @@
+import { resolveKimiHome } from "../kimi-home.js";
+import { probeProcessIdentity } from "../process-tree.js";
+import { recordRecoveryEvent, finishRecoveryRecord } from "../swarm-recovery.js";
 import { allocateJobId } from "../job-id.js";
-import { mkdir, readdir, stat, writeFile } from "node:fs/promises";
+import { mkdir, writeFile } from "node:fs/promises";
 import path from "node:path";
 import { createCliCancellationHandlers } from "../cli-cancellation.js";
 import { ResponseTimeoutError, runCliPromptWithBudget } from "../cli-client.js";
 import { deactivateSwarmSession, preactivateSwarmSession, requiresSwarmPreactivation, } from "../swarm-activation.js";
 import { executionPlanFromPersisted, observedExecutionFields, persistedExecutionPlanFields, prepareKimiExecutionPlan, } from "../kimi-engine.js";
 import { RuntimeError } from "../errors.js";
-import { captureWorktreePatch, createEphemeralWorktree, hasBornHead, isWorkingTreeDirty, pruneWorktrees, removeWorktree, resolveRepoIdentity, } from "../git.js";
+import { captureWorktreePatch, createEphemeralWorktree, hasBornHead, isWorkingTreeDirty, removeWorktree, resolveRepoIdentity, } from "../git.js";
 import { digestPrompt, markJobCancelled, markJobFailed, sweepStaleJobs } from "../jobs.js";
 import { JobStore } from "../job-store.js";
 import { classifyManagedCommandFailure } from "../kimi-errors.js";
 import { KIMI_SWARM_DEFAULT_BUDGET_MS } from "../kimi-timeouts.js";
 import { writeInvocationLogHeader } from "../logging.js";
 import { ensurePluginPaths, resolvePluginPaths } from "../paths.js";
-import { MAX_DURATION_MS, parseSwarmArgs } from "../parsing.js";
+import { parseSwarmArgs } from "../parsing.js";
 import { readArtifact, renderManagedJobOutput, writeArtifact } from "../render.js";
 import { hookRefusalDetails, hookRefusalRetryProtocol, maybeWarnHookMissing, verifyHookInstalled, } from "../hooks/install.js";
 import { assertCliResultSuccess, reassembleProseFromRecords, warnIfSessionIdMissing, warnHookEscapes } from "./cli-helpers.js";
@@ -25,27 +28,11 @@ import { buildKimiSessionTitle, syncKimiSessionTitle } from "../session-title.js
 // only the wall-clock hard bound. Read/write modes reuse review/rescue lineage,
 // and the mandatory PreToolUse hook enforces read-only policy or trusted-worktree
 // write confinement on the coordinator and every subagent. Write mode requires
-// kimi-code 0.18.0+. Exact 2.3.2 additionally activates a same-job session through
+// kimi-code 0.18.0+. Exact 2.3.2 and 2.3.3 additionally activates a same-job session through
 // a bounded loopback web lifecycle, then resumes it through the normal print path.
 const SWARM_SUMMARY_MAX = 120;
 const SWARM_AGENT_PROFILE = "<swarm>";
 const SWARM_WRITE_AGENT_PROFILE = "<swarm-write>";
-/**
- * Orphaned write-swarm worktrees older than this are reaped by the startup sweep.
- * Every normal run removes its own worktree in a finally; orphans can follow a
- * hard kill or failed cleanup.
- *
- * The sweep's only liveness signal is the worktree dir's mtime, which is frozen
- * at `git worktree add` time when a run edits only EXISTING files (in-place edits
- * never bump the containing dir's mtime). A live run can therefore look "idle" for
- * its entire wall-clock life. So the TTL MUST exceed the longest a run can live —
- * the hard `--budget` ceiling, `MAX_DURATION_MS` (24h) — or a concurrent run's
- * sweep could `git worktree remove --force` a still-live worktree out from under
- * its owner (ENOENT writes, lost patch). The +2h margin covers budget-expiry
- * teardown (AbortController fires → the finally removes the worktree well within
- * it). Orphans are still reaped, just after this longer, provably-safe delay.
- */
-export const WORKTREE_ORPHAN_TTL_MS = MAX_DURATION_MS + 2 * 60 * 60 * 1000;
 /**
  * Default concurrency setting when the user passes no `--max-concurrency`.
  * The plugin always exports KIMI_CODE_AGENT_SWARM_MAX_CONCURRENCY; kimi-code
@@ -254,7 +241,7 @@ async function runWriteSwarm(parsed, objective, context) {
         if (!(await hasBornHead(context.cwd))) {
             throw new RuntimeError("WRITE_SWARM_NO_HEAD", "/kimi:swarm --write needs a git repository with at least one commit (HEAD). This repo has an unborn HEAD; make an initial commit first.", "swarm.precondition");
         }
-        await sweepStaleWorktrees(paths, repoIdentity);
+        // Abandoned trees are recovery evidence, never age-based garbage.
         if (await isWorkingTreeDirty(context.cwd)) {
             context.stderr.write("[kimi-plugin-cc] /kimi:swarm --write bases its worktree on HEAD — your uncommitted changes are NOT visible to the swarm. Commit or stash them first if the swarm needs them.\n");
         }
@@ -362,37 +349,6 @@ function assertWriteSwarmPreconditions(repoIdentity) {
         throw new RuntimeError("WRITE_SWARM_NOT_A_REPO", "/kimi:swarm --write requires a git repository (it works in a throwaway worktree off HEAD). The current directory is not inside a git repo.", "swarm.precondition");
     }
 }
-/**
- * Reap orphaned write-swarm worktrees from prior crashed runs. Normal runs remove
- * their own worktree in a finally; orphans can follow hard kills or failed cleanup.
- * Prune stale admin entries (current repo), then force-remove on-disk worktree
- * dirs older than the TTL. Best-effort — never blocks the run.
- */
-async function sweepStaleWorktrees(paths, repoIdentity) {
-    await pruneWorktrees(repoIdentity.repoRoot);
-    // Only sweep THIS repo's namespaced subdir, so we never `rm` a worktree that a
-    // different repo still considers live (and never leave that repo a dangling
-    // admin entry). repoId is the realpath-stable hash from resolveRepoIdentity.
-    const repoWorktreesDir = path.join(paths.worktreesDir, repoIdentity.repoId);
-    const entries = await readdir(repoWorktreesDir, { withFileTypes: true }).catch(() => null);
-    if (entries === null)
-        return; // not created yet — nothing to sweep.
-    const now = Date.now();
-    for (const entry of entries) {
-        if (!entry.isDirectory() || !entry.name.startsWith("swarm-write-"))
-            continue;
-        const full = path.join(repoWorktreesDir, entry.name);
-        try {
-            const st = await stat(full);
-            if (now - st.mtimeMs > WORKTREE_ORPHAN_TTL_MS) {
-                await removeWorktree(repoIdentity.repoRoot, full);
-            }
-        }
-        catch {
-            // best-effort per entry
-        }
-    }
-}
 /**
  * Capture the worktree change set as a `.patch` artifact (write mode). Returns
  * the artifact path + size, or null on capture failure. Called on EVERY terminal
@@ -474,6 +430,7 @@ writeMode) {
     let cleanupAttempted = false;
     let cleanupFailure;
     let finalText;
+    let recovery;
     const commandLabel = writeMode ? "swarm-write" : "swarm";
     const recoveryCleanupTimeoutMs = 2_000;
     const cleanupSignal = (timeoutMs) => AbortSignal.any([
@@ -499,6 +456,14 @@ writeMode) {
     };
     try {
         const executionPlan = executionPlanFromPersisted(job);
+        if (requiresSwarmPreactivation(executionPlan.kimiVersion)) {
+            const owner = await probeProcessIdentity(process.pid);
+            recovery = { schema: 1, jobId: job.job_id, ownerPid: process.pid, owner: owner.identity ?? null,
+                kimiHome: resolveKimiHome(context.env, writeMode?.spawnCwd ?? job.cwd),
+                cwd: writeMode?.spawnCwd ?? job.cwd, workspaceRoot: writeMode?.workspaceRoot ?? null,
+                launches: [], session: null, sessionDirty: false, complete: false, patchPath: null };
+            store.saveSwarmRecovery(job.job_id, recovery);
+        }
         activationOptions = {
             cwd: writeMode?.spawnCwd ?? job.cwd,
             env: context.env,
@@ -511,6 +476,7 @@ writeMode) {
             swarmMaxConcurrency: maxConcurrency,
             trustedWorkspaceRoot: writeMode?.workspaceRoot,
             expectedVersion: executionPlan.kimiVersion ?? undefined,
+            onRecoveryEvent: recovery ? (event) => recordRecoveryEvent(store, recovery, event) : undefined,
         };
         if (requiresSwarmPreactivation(executionPlan.kimiVersion)) {
             store.updateRunningJob(job.job_id, { phase: "turn-running" });
@@ -577,6 +543,8 @@ writeMode) {
         if (promptBudgetMs <= 0)
             throw new ResponseTimeoutError(budgetMs, "swarm.prompt", commandLabel);
         store.updateRunningJob(job.job_id, { phase: "turn-running" });
+        if (recovery)
+            recordRecoveryEvent(store, recovery, { kind: "print-spawn" });
         const result = await runCliPromptWithBudget({
             cwd: activationOptions.cwd,
             env: context.env,
@@ -591,9 +559,17 @@ writeMode) {
             thinkingEffort: job.thinking_effort ?? undefined,
             resumeSessionId: preactivated?.sessionId,
             logPath: job.stream_log_path,
-            onSpawn: (pid) => { store.updateRunningJob(job.job_id, { kimi_pid: pid }); },
+            onSpawn: async (pid) => {
+                store.updateRunningJob(job.job_id, { kimi_pid: pid });
+                if (recovery) {
+                    const probe = await probeProcessIdentity(pid);
+                    recordRecoveryEvent(store, recovery, { kind: "print-processes", identities: probe.identity ? [probe.identity] : [], complete: probe.complete });
+                }
+            },
             signal: handlers.signal,
         }, promptBudgetMs, "swarm.prompt");
+        if (recovery)
+            recordRecoveryEvent(store, recovery, { kind: "print-exit" });
         store.updateRunningJob(job.job_id, observedExecutionFields(result));
         finalText = reassembleProseFromRecords(result.records);
         let resultError;
@@ -644,6 +620,8 @@ writeMode) {
             if (captured === null) {
                 throw new RuntimeError("SWARM_PATCH_CAPTURE_FAILED", `Swarm patch capture failed; worktree retained at ${writeMode.workspaceRoot}.`, "swarm.patch-capture", { details: { worktreePath: writeMode.workspaceRoot } });
             }
+            if (recovery)
+                recovery.patchPath = captured.path;
             finalText = `${renderWriteSwarmHeader(captured)}\n\n---\n\n${prose}`.trimEnd();
         }
         const rendered = renderManagedJobOutput(job, finalText);
@@ -662,6 +640,8 @@ writeMode) {
         if (handlers.cancelling) {
             throw new RuntimeError("SWARM_CANCELLED", "Swarm cancelled after artifact write.", "swarm.runtime");
         }
+        if (recovery)
+            finishRecoveryRecord(store, recovery);
         return (store.markCompleted(job.job_id, {
             summary: `[${writeMode ? "swarm-write" : "swarm"}] ${shorten(objective, SWARM_SUMMARY_MAX)}`,
             phase: "done",
@@ -689,6 +669,8 @@ writeMode) {
         if (writeMode && !(error instanceof RuntimeError && error.code === "SWARM_PATCH_CAPTURE_FAILED")) {
             const captured = await capturePatchArtifact(paths, job.job_id, writeMode.workspaceRoot, context.stderr, writeMode.onPatchCaptureFailure);
             captureFailed = captured === null;
+            if (recovery && captured)
+                recovery.patchPath = captured.path;
             if (captured && !captured.empty) {
                 context.stderr.write(`[kimi-plugin-cc] swarm-write run ended early; partial patch captured at ${captured.path}.\n`);
             }
@@ -745,7 +727,22 @@ writeMode) {
     }
     finally {
         handlers.dispose();
-        store.close();
+        try {
+            if (recovery)
+                finishRecoveryRecord(store, recovery);
+            if (recovery && !recovery.complete) {
+                writeMode?.onPatchCaptureFailure?.();
+                context.stderr.write(`[kimi-plugin-cc] recovery record retained: companion.sh recover-swarm ${job.job_id} (preview).\n`);
+            }
+        }
+        catch (error) {
+            // A failed final journal commit cannot authorize removal of recovery evidence.
+            writeMode?.onPatchCaptureFailure?.();
+            throw error;
+        }
+        finally {
+            store.close();
+        }
     }
 }
 function shorten(text, max) {

package/dist/companion.js

@@ -1,3 +1,4 @@
+import { runRecoverSwarm } from "./commands/recover-swarm.js";
 import { companionHelp } from "./help.js";
 import { runModelInventory } from "./commands/models.js";
 import { runCancel } from "./commands/cancel.js";
@@ -77,6 +78,9 @@ async function main(argv) {
         case "replay":
             context.stdout.write(await runReplay(rest, context));
             return;
+        case "recover-swarm":
+            context.stdout.write(await runRecoverSwarm(rest, context));
+            return;
         case "repair-sessions":
             context.stdout.write(await runRepairSessions(rest, context));
             return;
@@ -84,7 +88,7 @@ async function main(argv) {
             await runWorker(rest, context);
             return;
         default:
-            throw new RuntimeError("INVALID_COMMAND", `Unknown or missing companion subcommand: ${command ?? "<none>"}. Expected one of setup, review, task, ask, status, result, cancel, replay, repair-sessions.`, "companion");
+            throw new RuntimeError("INVALID_COMMAND", `Unknown or missing companion subcommand: ${command ?? "<none>"}. Expected one of setup, review, task, ask, status, result, cancel, replay, repair-sessions, recover-swarm.`, "companion");
     }
 }
 main(process.argv.slice(2)).catch((error) => {