Classify session/workspace create failures and persist field diagnostics.

Map Host internal errors on session.create and workspace.create to actionable codes, and write last-message-failure.json so field hosts can triage without guessing from HARNESS_SERVICE.

Co-authored-by: Cursor <cursoragent@cursor.com>

CHANGELOG.md

@@ -8,6 +8,9 @@ This file records the notable changes in each dsh-im release. Its format follows
 
 ### Fixed / 修复
 
+- `session.create` / `workspace.create` 的 Host `internal` 错误不再笼统显示 `HARNESS_SERVICE`：分别映射为 `SESSION_CREATE` / `WORKSPACE_UNAVAILABLE`，并写入 `%USERPROFILE%\\.dsh\\integrations\\dsh-im-ops\\last-message-failure.json`（含 method 与脱敏诊断），便于现场定位工作区路径问题。
+  Host `internal` failures on `session.create` / `workspace.create` no longer collapse to generic `HARNESS_SERVICE`: they map to `SESSION_CREATE` / `WORKSPACE_UNAVAILABLE`, and write `%USERPROFILE%\\.dsh\\integrations\\dsh-im-ops\\last-message-failure.json` (method + redacted diagnostics) for field triage of workspace path issues.
+
 - 发消息时若聊天绑定的 Session 不在机器人当前配置工作区，或 history 探针返回非「会话不存在」错误，自动解绑并按机器人工作区 / Agent Preset / 默认模型新建 Session，避免换工作区后幽灵绑定一直报 `HARNESS_SERVICE`（含重建后的 Host/Client 产物 `lib/`）。
   When a chat’s bound Session is outside the bot’s configured workspace, or the history probe fails with a non-missing Host error, IM now clears the binding and creates a new Session using the bot workspace / Agent Preset / default model, instead of sticking on a zombie binding that surfaces as `HARNESS_SERVICE` (includes rebuilt Host/Client `lib/` artifacts).
 

lib/client.js

@@ -571,7 +571,7 @@ var React26 = __toESM(require("react"), 1);
 // package.json
 var package_default = {
   name: "dsh-im-ops",
-  version: "4.9.1-ops.31",
+  version: "4.9.1-ops.32",
   description: "Ops fork of dsh-im (all channels kept). Based on @xmanrui/dsh-im@4.9.1 - access/session policies owned here.",
   keywords: [
     "deepseek-harness",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-im-ops",
-  "version": "4.9.1-ops.32",
+  "version": "4.9.1-ops.33",
   "description": "Ops fork of dsh-im (all channels kept). Based on @xmanrui/dsh-im@4.9.1 - access/session policies owned here.",
   "keywords": [
     "deepseek-harness",

src/channels/shared/i18n-en/shared-a.mjs

@@ -45,6 +45,8 @@ export default {
     'The bot and DeepSeek Harness use incompatible APIs. Ask an administrator to check the Harness address and update the related components.',
   'DeepSeek Harness 暂时无法完成请求，请稍后重试。':
     'DeepSeek Harness could not complete the request. Please try again later.',
+  '无法在当前工作区创建新会话。请检查机器人工作区是否为绝对路径、目录存在、且运行 Host 的账号可访问；改回可用工作区后重试。':
+    'Could not create a new Session in the current workspace. Check that the bot workspace is an absolute path, the directory exists, and the Host process can access it; then switch back to a working workspace and retry.',
   '等待模型回复超时，任务可能仍在运行。请先等待或发送 /stop，不要立即重复提交。':
     'Waiting for the model reply timed out and the task may still be running. Wait or send /stop before submitting it again.',
   '模型凭据缺失或已失效。请管理员检查模型配置后重试。':

src/channels/shared/message-failure.mjs

@@ -92,6 +92,8 @@ const FAILURE_MESSAGES = Object.freeze({
     '机器人与 DeepSeek Harness 的接口不兼容。请管理员检查 Harness 地址并更新相关版本。',
   HARNESS_SERVICE:
     'DeepSeek Harness 暂时无法完成请求，请稍后重试。',
+  SESSION_CREATE:
+    '无法在当前工作区创建新会话。请检查机器人工作区是否为绝对路径、目录存在、且运行 Host 的账号可访问；改回可用工作区后重试。',
   MODEL_REPLY_TIMEOUT:
     '等待模型回复超时，任务可能仍在运行。请先等待或发送 /stop，不要立即重复提交。',
   MODEL_AUTH:
@@ -159,8 +161,13 @@ function providerFailureCode(error) {
   return mapped;
 }
 
+function rpcMethod(error) {
+  return typeof error?.method === 'string' ? error.method.trim() : '';
+}
+
 function failureCode(error) {
   const code = typeof error?.code === 'string' ? error.code : '';
+  const method = rpcMethod(error);
   const providerCode = providerFailureCode(error);
   if (providerCode) return providerCode;
 
@@ -168,12 +175,12 @@ function failureCode(error) {
   if (isTransportProviderMessage(providerErrorMessage(error))) return 'MODEL_TRANSPORT';
 
   if (code === 'harness-connect-failed') {
-    return ['session.prompt', 'session.history'].includes(error?.method)
+    return ['session.prompt', 'session.history'].includes(method)
       ? 'HARNESS_RESULT_UNCERTAIN'
       : 'HARNESS_CONNECT';
   }
   if (code === 'harness-timeout') {
-    return ['session.prompt', 'session.history'].includes(error?.method)
+    return ['session.prompt', 'session.history'].includes(method)
       ? 'HARNESS_RESULT_UNCERTAIN'
       : 'HARNESS_TIMEOUT';
   }
@@ -187,8 +194,15 @@ function failureCode(error) {
     return 'HARNESS_PROTOCOL';
   }
   if (code === 'harness-turn-failed') return 'INTERNAL_UNKNOWN';
-  // Host RPC catch-all (common when session/preset/workspace ops fail without a typed code).
+  // Host RPC catch-all — refine by method so field ops see workspace vs create vs generic.
   if (code === 'internal' || ['harness-http-failed', 'harness-rpc-rejected'].includes(code)) {
+    if (method === 'workspace.create' || method === 'workspace.list') {
+      return 'WORKSPACE_UNAVAILABLE';
+    }
+    if (method === 'session.create') return 'SESSION_CREATE';
+    if (method.startsWith('agent-preset') || method.includes('preset')) {
+      return 'PRESET_UNAVAILABLE';
+    }
     return 'HARNESS_SERVICE';
   }
   if (code === 'model-empty-response') return 'MODEL_EMPTY_REPLY';
@@ -235,7 +249,20 @@ function inferredFailureReason(error) {
     ?? safeFailureReason(error?.name);
 }
 
-async function persistInternalUnknownDiagnostic(failure, error) {
+const PERSISTED_FAILURE_CODES = new Set([
+  'INTERNAL_UNKNOWN',
+  'HARNESS_SERVICE',
+  'SESSION_CREATE',
+  'WORKSPACE_UNAVAILABLE',
+  'PRESET_UNAVAILABLE',
+  'HARNESS_CONNECT',
+  'HARNESS_TIMEOUT',
+  'HARNESS_ACCESS',
+  'HARNESS_PROTOCOL',
+  'MODEL_TRANSPORT',
+]);
+
+async function persistMessageFailureDiagnostic(failure, error) {
   try {
     const dir = join(homedir(), '.dsh', 'integrations', 'dsh-im-ops');
     await mkdir(dir, { recursive: true });
@@ -250,11 +277,19 @@ async function persistInternalUnknownDiagnostic(failure, error) {
       method: typeof error?.method === 'string' ? error.method.slice(0, 80) : null,
       messages: diagnosticMessages(error).map((text) => text.slice(0, 240)),
     };
+    await writeFile(
+      join(dir, 'last-message-failure.json'),
+      `${JSON.stringify(payload, null, 2)}\n`,
+      { encoding: 'utf8', mode: 0o600 },
+    );
+    // Keep the legacy filename for INTERNAL_UNKNOWN so older runbooks still work.
+    if (failure.code === 'INTERNAL_UNKNOWN') {
       await writeFile(
         join(dir, 'last-internal-unknown.json'),
         `${JSON.stringify(payload, null, 2)}\n`,
         { encoding: 'utf8', mode: 0o600 },
       );
+    }
   } catch {
     // Diagnostics must never mask the user-facing failure path.
   }
@@ -285,9 +320,10 @@ export function classifyMessageFailure(error, {
     && userMessage.trim()
     ? 'INPUT_INVALID'
     : classifiedCode;
+  const methodReason = safeFailureReason(rpcMethod(error)?.replaceAll('.', '_'));
   const safeReason = code === 'INTERNAL_UNKNOWN'
     ? (explicitReason ?? inferredFailureReason(error) ?? code)
-    : (explicitReason ?? code);
+    : (explicitReason ?? methodReason ?? code);
   return Object.freeze({
     code,
     reason: safeReason,
@@ -306,8 +342,8 @@ export function messageFailureText(failure) {
 export function setLastMessageFailure(status, error, options) {
   const failure = classifyMessageFailure(error, options);
   status.lastMessageError = failure;
-  if (failure.code === 'INTERNAL_UNKNOWN') {
-    void persistInternalUnknownDiagnostic(failure, error);
+  if (PERSISTED_FAILURE_CODES.has(failure.code)) {
+    void persistMessageFailureDiagnostic(failure, error);
   }
   return failure;
 }

test/message-failure.test.mjs

@@ -88,6 +88,12 @@ test('PI_AI_ERROR with TLS or transport wording maps to MODEL_TRANSPORT', () =>
 test('RPC internal and UNKNOWN provider codes no longer fall through to INTERNAL_UNKNOWN', () => {
   assert.equal(classifyMessageFailure({
     code: 'internal', method: 'session.create',
+  }, options).code, 'SESSION_CREATE');
+  assert.equal(classifyMessageFailure({
+    code: 'internal', method: 'workspace.create',
+  }, options).code, 'WORKSPACE_UNAVAILABLE');
+  assert.equal(classifyMessageFailure({
+    code: 'internal', method: 'session.prompt',
   }, options).code, 'HARNESS_SERVICE');
   assert.equal(classifyMessageFailure({
     code: 'harness-turn-failed', providerCode: 'UNKNOWN',
@@ -102,6 +108,15 @@ test('RPC internal and UNKNOWN provider codes no longer fall through to INTERNAL
   assert.equal(plain.reason, 'ERROR');
 });
 
+test('session.create failures tell operators to check the bot workspace path', () => {
+  const failure = classifyMessageFailure({
+    code: 'internal', method: 'session.create',
+  }, options);
+  assert.equal(failure.code, 'SESSION_CREATE');
+  assert.equal(failure.reason, 'SESSION_CREATE');
+  assert.match(failure.message, /工作区/);
+});
+
 test('message failure text contains a safe code and traceable reference', () => {
   const failure = classifyMessageFailure(new Error('secret-shaped internal detail'), options);
   assert.deepEqual(failure, {