@ljwei-stak/dsh-model-router 0.19.0 → 0.20.1

package/.dsh-plugin/shared/official-tools-remote.mjs

@@ -127,6 +127,7 @@ const rerunRequestCodec = strictCodec(`${OFFICIAL_TOOLS_REMOTE_PACKAGE}#RerunReq
     ...(provider ? { provider, model } : {}),
     confirmOverBudget: request.confirmOverBudget === true,
     confirmWrite: request.confirmWrite === true,
+    ...(request.confirmUnsandboxed === undefined ? {} : { confirmUnsandboxed: request.confirmUnsandboxed === true }),
     ...(['api', 'subscription', 'cancel'].includes(request.subscriptionChoice) ? { subscriptionChoice: request.subscriptionChoice } : {}),
   }
 })
@@ -152,11 +153,22 @@ export function parseRunRequest(value) {
     throw new TypeError('budgetUsd must be a non-negative number')
   }
   const confirmed = Array.isArray(request.confirmedReasons) ? request.confirmedReasons : []
+  if (request.executor !== undefined && !['routed', 'official-team'].includes(request.executor)) throw new TypeError('executor is invalid')
+  if (request.executionMode !== undefined && !['read-only', 'workspace-write'].includes(request.executionMode)) throw new TypeError('executionMode is invalid')
+  if (request.executionEnvironment !== undefined && !['harness', 'cli-native'].includes(request.executionEnvironment)) throw new TypeError('executionEnvironment is invalid')
+  const officialTeam = request.executor === 'official-team'
+  if (officialTeam && (request.planMode !== 'team' || provider || model)) throw new TypeError('official-team requires a team plan without an explicit route')
+  if (!officialTeam && request.executionMode === 'workspace-write') throw new TypeError('workspace-write requires official-team')
+  if (request.executionEnvironment === 'cli-native' && (!officialTeam || request.executionMode !== 'workspace-write')) {
+    throw new TypeError('cli-native requires official-team with workspace-write')
+  }
   if (confirmed.length > RUN_CONFIRMATION_CODES.length || confirmed.some(code => !RUN_CONFIRMATION_CODES.includes(code))) {
     throw new TypeError('confirmedReasons contains an unknown reason')
   }
   return {
     task: request.task,
+    ...(officialTeam ? { executor: 'official-team', executionMode: request.executionMode ?? 'read-only',
+      ...(request.executionEnvironment === undefined ? {} : { executionEnvironment: request.executionEnvironment }) } : {}),
     ...(provider ? { provider, model } : {}),
     ...(['single', 'team'].includes(request.planMode) ? { planMode: request.planMode } : {}),
     ...(['economy', 'balanced', 'quality'].includes(request.preset) ? { preset: request.preset } : {}),

package/.dsh-plugin/shared/official-tools-runtime.mjs

@@ -850,6 +850,22 @@ export async function probeToolWith(tool, runner, {
   return probe
 }
 
+/** Reuse the concrete entry discovered by the installation probe for login checks. */
+export async function loginLaunchForProbe(probe, {
+  workspace = process.cwd(),
+  fileExists = async path => (await stat(path).catch(() => null))?.isFile() === true,
+  findNode = async cwd => (await import('./official-tool-executor.mjs')).findNodeExecutable(cwd),
+} = {}) {
+  if (probe?.installed !== true || !getOfficialTool(probe.id)) return null
+  const entry = probe.packageEntry || probe.executablePath
+  if (typeof entry !== 'string' || !isAbsolute(entry) || !(await fileExists(entry))) return null
+  if (/\.(?:mjs|cjs|js)$/i.test(entry)) {
+    const node = await findNode(workspace)
+    return node ? { file: node, args: [entry], useShell: false } : null
+  }
+  return { file: entry, args: [], useShell: IS_WINDOWS && /\.(?:cmd|bat)$/i.test(entry) }
+}
+
 async function selectedPackageEntries(tool, { plan, preferences: selectedPreferences }) {
   const { findGlobalPackageEntries, officialToolEntrySelection } = await import('./official-tool-executor.mjs')
   const selection = officialToolEntrySelection(tool.id, { preferences: selectedPreferences, plan })
@@ -888,7 +904,12 @@ async function probeUncached(tool, runner, { selectedPreferences, plan, fileExis
         ...selectedPreferences,
         allowScriptInstall: installMethodFor(tool, selectedPreferences.methods?.[tool.id])?.kind === 'script' ? true : selectedPreferences.allowScriptInstall,
       }, currentInstallOptions())
-    } catch { /* a blocked saved method still allows normal detection */ }
+    } catch (error) {
+      if (usesConfiguredInstallPlan(tool.id, selectedPreferences)) {
+        return { id: tool.id, installed: false, version: null, status: 'probe-failed',
+          detail: `无法检查所选安装方式或目录：${String(error?.message ?? error).slice(0, 200)}` }
+      }
+    }
   }
   // npm and pnpm can write the same shim into a unified bin directory. Its
   // banner belongs to whichever manager wrote it last, not the selected copy.

package/.dsh-plugin/shared/run-ledger.mjs

@@ -12,12 +12,30 @@ export const MAX_QUALITY_BIAS = 0.04
 const finite = value => typeof value === 'number' && Number.isFinite(value)
 const routeKey = (provider, model) => `${String(provider ?? '')}\u0000${String(model ?? '')}`
 
+function modelIdentity(value) {
+  if (typeof value === 'string' && value.trim()) return value.trim()
+  if (value && typeof value === 'object' && !Array.isArray(value)
+    && typeof value.providerId === 'string' && value.providerId
+    && typeof value.modelId === 'string' && value.modelId) return `${value.providerId}/${value.modelId}`
+  return null
+}
+
 /** USD cost of one package. Prefers a CLI-reported cost, else usage × configured price. */
 export function actualCost(result, pricing) {
   if (result && Object.hasOwn(result, 'reportedCostUsd') && finite(result.reportedCostUsd) && result.reportedCostUsd >= 0) {
     return { costUsd: result.reportedCostUsd, costSource: 'cli-reported' }
   }
   const usage = result && Object.hasOwn(result, 'usage') ? result.usage : null
+  if (usage && result.usageIncomplete === true) return { costUsd: null, costSource: 'usage-incomplete' }
+  if (usage && result.channel === 'official-cli') {
+    const actualModel = modelIdentity(result.actualModel) ?? modelIdentity(result.reportedModel)
+    if (!actualModel) return { costUsd: null, costSource: 'model-unknown' }
+    const pricedModel = pricing?.model ?? result.model
+    const provider = pricing?.provider ?? result.provider
+    const matches = [pricedModel, pricing?.cliModel, provider && pricedModel ? `${provider}/${pricedModel}` : null]
+      .some(value => typeof value === 'string' && value === actualModel)
+    if (!matches) return { costUsd: null, costSource: 'model-price-unverified' }
+  }
   if (!usage || !pricing || !Object.hasOwn(pricing, 'input') || !Object.hasOwn(pricing, 'output')
     || pricing.input === undefined || pricing.input === null || pricing.output === undefined || pricing.output === null) {
     return { costUsd: null, costSource: usage ? 'price-missing' : 'usage-missing' }
@@ -268,12 +286,25 @@ function storedPackage(planned, result, pricing, finishedAt, billingFor = billin
     channel: result?.channel ?? null,
     toolId: result?.toolId ?? null,
     credentialSource: result?.credentialSource ?? null,
-    ...(result?.actualModel ? { actualModel: String(result.actualModel) } : {}),
+    ...(result?.channel === 'official-cli' ? {
+      requestedCliModel: result.requestedCliModel ?? null,
+      actualModel: modelIdentity(result.actualModel) ?? modelIdentity(result.reportedModel),
+      modelNotice: result.modelNotice ?? null,
+    } : {}),
+    ...(result?.channel !== 'official-cli' && result?.actualModel ? { actualModel: String(result.actualModel) } : {}),
+    ...(result?.reportedModel !== undefined ? { reportedModel: result.reportedModel } : {}),
+    ...(typeof result?.source === 'string' ? { source: result.source } : {}),
+    ...(typeof result?.sandboxEnforcement === 'string' ? { sandboxEnforcement: result.sandboxEnforcement } : {}),
+    ...(result?.usageIncomplete === true ? { usageIncomplete: true } : {}),
+    ...(typeof result?.executionEnvironment === 'string' ? { executionEnvironment: result.executionEnvironment } : {}),
+    ...(finite(result?.reportedCostUsd) && result.reportedCostUsd >= 0 ? { reportedCostUsd: result.reportedCostUsd } : {}),
     fallback: result?.fallback ?? null,
     ...(result?.billingMode ? { billingMode: String(result.billingMode) } : {}),
     ...(result?.billingSwitch ? { billingSwitch: { ...result.billingSwitch } } : {}),
     ...(result?.subscriptionRoute ? { subscriptionRoute: { provider: String(result.subscriptionRoute.provider ?? ''), model: String(result.subscriptionRoute.model ?? '') } } : {}),
     error: result?.ok ? null : (result?.error ?? null),
+    ...(!result?.ok && result?.diagnostic ? { diagnostic: String(result.diagnostic).slice(0, 2_000) } : {}),
+    ...(Number.isInteger(result?.exitCode) ? { exitCode: result.exitCode } : {}),
     answer: String(result?.answer ?? '').slice(0, MAX_STORED_ANSWER),
     answerTruncated: String(result?.answer ?? '').length > MAX_STORED_ANSWER,
     usage: result?.usage ?? null,
@@ -375,12 +406,21 @@ export function teamExecutionResults(plan, execution) {
     cancelled: result.status === 'cancelled',
     provider: result.provider,
     model: result.recommendedModel,
+    requestedCliModel: result.requestedCliModel ?? null,
     actualModel: result.actualModel ?? null,
+    ...(result.reportedModel !== undefined ? { reportedModel: result.reportedModel } : {}),
+    modelNotice: result.modelNotice ?? null,
+    ...(typeof result.source === 'string' ? { source: result.source } : {}),
+    ...(typeof result.sandboxEnforcement === 'string' ? { sandboxEnforcement: result.sandboxEnforcement } : {}),
+    ...(result.usageIncomplete === true ? { usageIncomplete: true } : {}),
+    ...(typeof result.executionEnvironment === 'string' ? { executionEnvironment: result.executionEnvironment } : {}),
     channel: 'official-cli',
     toolId: result.toolId ?? null,
     answer: String(result.finalText ?? ''),
     error: result.status === 'succeeded' ? null : String(result.error ?? (TEAM_FAILURE.has(result.status) ? result.status : '官方 CLI 未成功完成。')),
     stderrTail: result.outputTail ?? null,
+    diagnostic: result.outputTail ?? null,
+    exitCode: result.exitCode ?? null,
     ...(result.usage ? { usage: result.usage } : {}),
     ...(finite(result.reportedCostUsd) ? { reportedCostUsd: result.reportedCostUsd } : {}),
   }))
@@ -421,9 +461,18 @@ export function buildToolRunRecord({ toolId, toolLabel, provider, model, cliMode
       id: packageId, ok: succeeded, cancelled: result?.status === 'cancelled',
       provider: plan.directRoute.provider, model: plan.directRoute.model,
       channel: 'official-cli', toolId,
-      actualModel: result?.reportedModel && typeof result.reportedModel === 'string' ? result.reportedModel : null,
+      requestedCliModel: Object.hasOwn(result ?? {}, 'requestedModel') ? result.requestedModel : cliModel,
+      actualModel: modelIdentity(result?.actualModel) ?? modelIdentity(result?.reportedModel),
+      ...(result?.reportedModel !== undefined ? { reportedModel: result.reportedModel } : {}),
+      modelNotice: result?.modelNotice ?? null,
+      ...(typeof result?.source === 'string' ? { source: result.source } : {}),
+      ...(typeof result?.sandboxEnforcement === 'string' ? { sandboxEnforcement: result.sandboxEnforcement } : {}),
+      ...(result?.usageIncomplete === true ? { usageIncomplete: true } : {}),
       answer: String(result?.finalText ?? ''),
       error: succeeded ? null : String(result?.error ?? result?.reason ?? result?.status ?? '官方 CLI 未成功完成。'),
+      ...(!succeeded && typeof result?.diagnostic === 'string' && result.diagnostic.trim()
+        ? { diagnostic: result.diagnostic } : {}),
+      ...(Number.isInteger(result?.exitCode) ? { exitCode: result.exitCode } : {}),
       ...(result?.usage ? { usage: result.usage } : {}),
       ...(finite(result?.reportedCostUsd) ? { reportedCostUsd: result.reportedCostUsd } : {}),
     }],

package/.dsh-plugin/shared/task-executors.mjs

@@ -36,7 +36,7 @@ const BASE_ENV_KEYS = Object.freeze([
   'PATH', 'Path', 'PATHEXT', 'SystemRoot', 'windir', 'ComSpec',
   'HOME', 'USERPROFILE', 'HOMEDRIVE', 'HOMEPATH', 'APPDATA', 'LOCALAPPDATA',
   'PROGRAMDATA', 'TEMP', 'TMP', 'TMPDIR', 'LANG', 'LC_ALL', 'TERM',
-  'HTTP_PROXY', 'HTTPS_PROXY', 'NO_PROXY', 'http_proxy', 'https_proxy', 'no_proxy',
+  'HTTP_PROXY', 'HTTPS_PROXY', 'ALL_PROXY', 'NO_PROXY', 'http_proxy', 'https_proxy', 'all_proxy', 'no_proxy',
   'SSL_CERT_FILE', 'NODE_EXTRA_CA_CERTS',
 ])
 
@@ -50,7 +50,9 @@ export const TASK_ADAPTERS = Object.freeze([
     executable: 'claude',
     portable: true,
     apiKeyEnv: 'ANTHROPIC_API_KEY',
-    sessionEnv: Object.freeze(['CLAUDE_CODE_OAUTH_TOKEN', 'CLAUDE_CONFIG_DIR']),
+    sessionEnv: Object.freeze(['CLAUDE_CODE_OAUTH_TOKEN', 'CLAUDE_CONFIG_DIR', 'ANTHROPIC_AUTH_TOKEN',
+      'ANTHROPIC_BASE_URL', 'ANTHROPIC_MODEL', 'ANTHROPIC_DEFAULT_HAIKU_MODEL', 'ANTHROPIC_DEFAULT_SONNET_MODEL',
+      'ANTHROPIC_DEFAULT_OPUS_MODEL', 'ANTHROPIC_SMALL_FAST_MODEL']),
     format: 'claude-json',
     buildArgs(modelId) {
       return [
@@ -193,17 +195,22 @@ async function checkedWorkspace(workspace) {
 }
 
 function cliModelFor(route, adapter) {
-  const explicit = checkedModel(route?.cliModel)
-  if (explicit) return explicit
-  if (adapter?.id === 'claude-code' || adapter?.id === 'codex' || adapter?.id === 'gemini') {
-    return checkedModel(route?.model)
+  // Catalog IDs are routing recommendations, not vendor CLI aliases. Omit
+  // the model argument unless the route explicitly binds a CLI model.
+  const value = route?.cliModel
+  if (value === undefined || value === null || value === '') return null
+  const explicit = checkedModel(value)
+  if (!explicit) throw new TypeError('cliModel must be a bounded model identifier')
+  if ((adapter?.id === 'minimax-code' || adapter?.id === 'mimo-code') && !explicit.includes('/')) {
+    throw new TypeError(`${adapter.id} cliModel requires provider/model`)
   }
-  return null
+  if (adapter?.id === 'zcode') throw new TypeError('ZCode CLI does not support a per-call cliModel')
+  return explicit
 }
 
 /** Variables that switch a CLI from its account login to API-key billing. */
 const API_KEY_VARIABLES = Object.freeze({
-  'claude-code': ['ANTHROPIC_API_KEY'],
+  'claude-code': ['ANTHROPIC_API_KEY', 'ANTHROPIC_AUTH_TOKEN'],
   codex: ['OPENAI_API_KEY', 'CODEX_API_KEY'],
   gemini: ['GEMINI_API_KEY', 'GOOGLE_API_KEY'],
 })
@@ -324,7 +331,7 @@ function parseAdapterOutput(format, stdout) {
 const DIAGNOSTIC_CHARS = 400
 
 /** Shorten and scrub a CLI diagnostic before it reaches the user or the run ledger. */
-export function redactDiagnostic(value, secret = '') {
+export function redactDiagnostic(value, secret = '', maxChars = DIAGNOSTIC_CHARS) {
   const scrubbed = redact(String(value ?? ''), secret)
     .replace(/sk-[A-Za-z0-9_-]{8,}/g, 'sk-[redacted]')
     .replace(/(Bearer\s+)[A-Za-z0-9._~+/=-]{8,}/gi, '$1[redacted]')
@@ -332,7 +339,10 @@ export function redactDiagnostic(value, secret = '') {
     .replace(/\b(api[_-]?key|token|secret|password)(["'\s:=]+)[^\s"',}]{6,}/gi, '$1$2[redacted]')
     .replace(/\s+/g, ' ')
     .trim()
-  return scrubbed.length > DIAGNOSTIC_CHARS ? `${scrubbed.slice(0, DIAGNOSTIC_CHARS)}…` : scrubbed
+  const limit = Number.isInteger(maxChars) ? Math.max(100, Math.min(4_000, maxChars)) : DIAGNOSTIC_CHARS
+  if (scrubbed.length <= limit) return scrubbed
+  const head = Math.floor(limit / 3)
+  return `${scrubbed.slice(0, head)}…${scrubbed.slice(-(limit - head - 1))}`
 }
 
 function jsonLines(stdout) {
@@ -389,7 +399,9 @@ export function usageFromOutput(format, stdout = '') {
       const result = JSON.parse(stdout)
       const usage = result?.usage
       if (!usage || typeof usage !== 'object') return null
-      const reported = Number(result.total_cost_usd)
+      const reported = result.total_cost_usd
+      const models = result.modelUsage && typeof result.modelUsage === 'object' && !Array.isArray(result.modelUsage)
+        ? Object.keys(result.modelUsage) : []
       return {
         usage: {
           inputTokens: count(usage.input_tokens),
@@ -397,7 +409,8 @@ export function usageFromOutput(format, stdout = '') {
           cacheReadTokens: count(usage.cache_read_input_tokens),
           cacheWriteTokens: count(usage.cache_creation_input_tokens),
         },
-        ...(Number.isFinite(reported) && reported > 0 ? { reportedCostUsd: reported } : {}),
+        ...(Number.isFinite(reported) && reported >= 0 ? { reportedCostUsd: reported } : {}),
+        ...(models.length === 1 && /^[A-Za-z0-9][A-Za-z0-9._:/\[\]-]{0,199}$/.test(models[0]) ? { reportedModel: models[0] } : {}),
       }
     }
     if (format === 'codex-jsonl') {
@@ -442,26 +455,27 @@ export function usageFromEvents(format, value) {
   if (format === 'grok-end') {
     const usage = value?.usage
     if (!usage || typeof usage !== 'object') return null
-    const reported = Number(value.total_cost_usd)
+    const reported = value.total_cost_usd
     return {
       usage: {
         inputTokens: count(usage.input_tokens), outputTokens: count(usage.output_tokens),
         cacheReadTokens: count(usage.cache_read_input_tokens), cacheWriteTokens: count(usage.cache_creation_input_tokens),
       },
-      ...(value.cost_is_partial !== true && Number.isFinite(reported) && reported > 0 ? { reportedCostUsd: reported } : {}),
+      ...(value.cost_is_partial !== true && Number.isFinite(reported) && reported >= 0 ? { reportedCostUsd: reported } : {}),
     }
   }
   if (format === 'mimo-steps') {
     const parts = (Array.isArray(value) ? value : []).filter(part => part?.tokens && typeof part.tokens === 'object')
     if (!parts.length) return null
     const sum = pick => parts.reduce((total, part) => total + count(pick(part)), 0)
-    const cost = parts.reduce((total, part) => total + (Number.isFinite(part.cost) && part.cost > 0 ? part.cost : 0), 0)
+    const costKnown = parts.every(part => Number.isFinite(part.cost) && part.cost >= 0)
+    const cost = costKnown ? parts.reduce((total, part) => total + part.cost, 0) : null
     return {
       usage: {
         inputTokens: sum(part => part.tokens.input), outputTokens: sum(part => part.tokens.output) + sum(part => part.tokens.reasoning),
         cacheReadTokens: sum(part => part.tokens.cache?.read), cacheWriteTokens: sum(part => part.tokens.cache?.write),
       },
-      ...(cost > 0 ? { reportedCostUsd: Number(cost.toFixed(8)) } : {}),
+      ...(costKnown ? { reportedCostUsd: Number(cost.toFixed(8)) } : {}),
     }
   }
   if (format === 'minimax-result') {
@@ -648,11 +662,30 @@ async function useApi(apiFallback, { route, task, signal, reason, adapter, prefe
   }
 }
 
-function officialSuccess({ route, adapter, preference, answer, exitCode, version, credentialSource, usage = null }) {
+function reportedCliModel(value) {
+  if (typeof value === 'string' && value.trim()) return value.trim()
+  if (value && typeof value === 'object' && !Array.isArray(value)
+    && typeof value.providerId === 'string' && value.providerId
+    && typeof value.modelId === 'string' && value.modelId) {
+    return `${value.providerId}/${value.modelId}`
+  }
+  return null
+}
+
+function officialSuccess({ route, adapter, preference, answer, exitCode, version, credentialSource, usage = null,
+  requestedCliModel = null, reportedModel = null, modelNotice = null, source, sandboxEnforcement }) {
+  const actualModel = reportedCliModel(reportedModel)
   return {
     ok: true,
     provider: route.provider,
     model: route.model,
+    recommendedModel: route.model,
+    requestedCliModel,
+    actualModel,
+    ...(reportedModel !== null && reportedModel !== undefined ? { reportedModel } : {}),
+    modelNotice: modelNotice ?? (!requestedCliModel
+      ? '本次使用厂商 CLI 已配置的默认模型；Harness 推荐模型 ID 未作为 CLI 模型参数。'
+      : actualModel ? null : '该 CLI 未返回可核验的实际模型 ID；请以厂商运行记录核对。'),
     preference,
     channel: 'official-cli',
     toolId: adapter.id,
@@ -664,8 +697,11 @@ function officialSuccess({ route, adapter, preference, answer, exitCode, version
     exitCode,
     version: version ?? null,
     credentialSource,
+    ...(typeof source === 'string' ? { source } : {}),
+    ...(typeof sandboxEnforcement === 'string' ? { sandboxEnforcement } : {}),
     ...(usage?.usage ? { usage: usage.usage } : {}),
-    ...(usage?.reportedCostUsd ? { reportedCostUsd: usage.reportedCostUsd } : {}),
+    ...(Number.isFinite(usage?.reportedCostUsd) && usage.reportedCostUsd >= 0 ? { reportedCostUsd: usage.reportedCostUsd } : {}),
+    ...(usage?.usageIncomplete === true ? { usageIncomplete: true } : {}),
   }
 }
 
@@ -758,6 +794,13 @@ export async function executeAssignedTask({
       return officialSuccess({
         route, adapter, preference, answer: redact(verified.finalText, secret),
         exitCode: verified.exitCode ?? 0, version: null, credentialSource,
+        requestedCliModel: Object.hasOwn(verified, 'requestedModel') ? verified.requestedModel : modelId,
+        reportedModel: verified.reportedModel,
+        modelNotice: verified.modelNotice,
+        source: verified.source,
+        sandboxEnforcement: verified.sandboxEnforcement,
+        usage: { usage: verified.usage, reportedCostUsd: verified.reportedCostUsd,
+          usageIncomplete: verified.usageIncomplete },
       })
     }
     if (verified && verified.status !== 'unsupported') {
@@ -833,6 +876,7 @@ export async function executeAssignedTask({
     route, adapter, preference, answer, exitCode: run.exitCode,
     version: versionFromBanner(`${probe.stdout}\n${probe.stderr}`),
     credentialSource,
+    requestedCliModel: modelId,
     usage: usageFromOutput(adapter.format, run.stdout),
   })
 }

package/.dsh-plugin/shared/tool-health.mjs

@@ -14,6 +14,7 @@
  */
 import { OFFICIAL_TOOLS, getOfficialTool } from './official-tool-registry.mjs'
 import { compareReleaseVersions } from './latest-versions.mjs'
+import { loginLaunchForProbe } from './official-tools-runtime.mjs'
 
 export const LOGIN_CHECK_TIMEOUT_MS = 5_000
 export const HEALTH_CACHE_MS = 10 * 60_000
@@ -34,7 +35,7 @@ export const LOGIN_GUIDES = Object.freeze({
 
 /** Environment variable names whose presence means a CLI can authenticate without a session. */
 const KEY_ENV = Object.freeze({
-  'claude-code': ['ANTHROPIC_API_KEY', 'CLAUDE_CODE_OAUTH_TOKEN'],
+  'claude-code': ['ANTHROPIC_API_KEY', 'ANTHROPIC_AUTH_TOKEN', 'CLAUDE_CODE_OAUTH_TOKEN'],
   codex: ['OPENAI_API_KEY', 'CODEX_API_KEY'],
   gemini: ['GEMINI_API_KEY', 'GOOGLE_API_KEY'],
   'kimi-code': ['KIMI_API_KEY', 'MOONSHOT_API_KEY'],
@@ -92,7 +93,7 @@ function grokHomeHint(env) {
 
 /** Variables that make the CLI bill an API account rather than a subscription login. */
 const API_KEY_ENV = Object.freeze({
-  'claude-code': ['ANTHROPIC_API_KEY'],
+  'claude-code': ['ANTHROPIC_API_KEY', 'ANTHROPIC_AUTH_TOKEN'],
   codex: ['OPENAI_API_KEY', 'CODEX_API_KEY'],
   gemini: ['GEMINI_API_KEY', 'GOOGLE_API_KEY'],
   'kimi-code': ['KIMI_API_KEY', 'MOONSHOT_API_KEY'],
@@ -151,14 +152,30 @@ function guide(toolId) {
  * Login state for one installed tool: 'logged-in', 'logged-out', or 'unknown'.
  * `runner(executable, args, { timeoutMs })` resolves { ok, code, stdout, stderr, timedOut }.
  */
-export async function checkLogin(toolId, { runner, env = process.env, exists = async () => false, home = '', readAuthState = async () => null } = {}) {
+export async function checkLogin(toolId, {
+  runner, env = process.env, exists = async () => false, home = '', readAuthState = async () => null,
+  probe, resolveLoginLaunch = loginLaunchForProbe,
+} = {}) {
   const keyNames = (KEY_ENV[toolId] ?? []).filter(name => typeof env?.[name] === 'string' && env[name].trim())
   const viaKey = keyNames.length > 0 ? { state: 'logged-in', detail: `检测到环境变量 ${keyNames.join('、')}（未验证有效性）。`, source: 'environment',
     billing: apiKeyEnvPresent(toolId, env) ? 'api-key' : 'subscription',
     accountLogin: keyNames.some(name => !(API_KEY_ENV[toolId] ?? []).includes(name)) } : null
   const keyBilling = apiKeyEnvPresent(toolId, env) ? 'api-key' : null
+  // A health report must inspect the same npm/pnpm/native copy whose version
+  // was shown. Never replace a missing selected entry with a bare PATH command.
+  const statusCommand = async (command, args) => {
+    const launch = probe === undefined
+      ? { file: command, args: [] }
+      : await resolveLoginLaunch(probe)
+    if (!launch) return null
+    return runner(launch.file, [...launch.args, ...args], {
+      timeoutMs: LOGIN_CHECK_TIMEOUT_MS, env: sessionEnvironment(toolId, env),
+      ...(launch.useShell === undefined ? {} : { useShell: launch.useShell }),
+    })
+  }
   if (toolId === 'claude-code' && typeof runner === 'function') {
-    const outcome = await runner('claude', ['auth', 'status', '--json'], { timeoutMs: LOGIN_CHECK_TIMEOUT_MS, env: sessionEnvironment(toolId, env) })
+    const outcome = await statusCommand('claude', ['auth', 'status', '--json'])
+    if (!outcome) return viaKey ?? { state: 'unknown', detail: '所选安装副本的登录检测入口不可用；未使用 PATH 中的其他 Claude Code。' }
     if (outcome.timedOut) return viaKey ?? { state: 'unknown', detail: '登录状态检测超时。' }
     try {
       const status = JSON.parse(outcome.stdout)
@@ -170,7 +187,8 @@ export async function checkLogin(toolId, { runner, env = process.env, exists = a
     return viaKey ?? { state: 'unknown', detail: '无法解析 claude auth status 输出。' }
   }
   if (toolId === 'codex' && typeof runner === 'function') {
-    const outcome = await runner('codex', ['login', 'status'], { timeoutMs: LOGIN_CHECK_TIMEOUT_MS, env: sessionEnvironment(toolId, env) })
+    const outcome = await statusCommand('codex', ['login', 'status'])
+    if (!outcome) return viaKey ?? { state: 'unknown', detail: '所选安装副本的登录检测入口不可用；未使用 PATH 中的其他 Codex。' }
     if (outcome.timedOut) return viaKey ?? { state: 'unknown', detail: '登录状态检测超时。' }
     const output = `${outcome.stdout}\n${outcome.stderr}`
     if (/not logged in/i.test(output)) return viaKey ?? { state: 'logged-out', detail: 'codex login status 报告未登录。', source: 'cli' }
@@ -234,7 +252,7 @@ export async function checkToolHealth(probe, options = {}) {
   const tool = getOfficialTool(probe?.id)
   if (!tool) throw new TypeError('unknown official tool')
   const installed = probe.installed === true
-  const login = installed ? await checkLogin(tool.id, options) : { state: 'unknown', detail: probe.status === 'not-installed' && /检测到已安装/.test(String(probe.detail ?? '')) ? String(probe.detail) : '尚未安装。' }
+  const login = installed ? await checkLogin(tool.id, { ...options, probe }) : { state: 'unknown', detail: probe.status === 'not-installed' && /检测到已安装/.test(String(probe.detail ?? '')) ? String(probe.detail) : '尚未安装。' }
   const { command, steps } = guide(tool.id)
   return {
     id: tool.id,

package/CHANGELOG.md

@@ -1,5 +1,31 @@
 # Changelog
 
+## 0.20.1 — 2026-10-16
+
+- Expand the English and Simplified Chinese READMEs together: architecture and execution contracts, workbench/session request examples, platform capabilities and executable trust, sequential team lifecycle, model identity, worktree integration/retry, accounting formulas, routing weights/floors/search limits, persistence and verification layers. Add matching navigation anchors and record the reviewed translation pair in `README.i18n.yaml`.
+- Correct stale claims about Windows Codex editable support and the Harness/native permission boundary. Distinguish registry-file consistency from npm provenance and clarify installer-marker and ZCode first-seen script trust.
+- Forward explicit unsandboxed confirmation through the workbench retry dispatcher, allowing an approved native CLI team retry to reach the existing execution gate. Missing confirmation remains blocked. Add focused regression coverage; no new model calls or billing claims. The dated 0.20.0 real-execution report remains unchanged.
+
+Validation: strict frozen pnpm 10.34.6 install, real client build and generated-client consistency passed; 834 tests, 831 passed, 3 conditional skips, no failures. The new gateway/Host retry regression fails before the forwarding fix and passes afterward. Bilingual checks cover section anchors, table shape, local links and protocol/formula/command examples.
+
+## 0.20.0 — 2026-10-15
+
+Fully quit Harness, including its tray or WebUI Host process, and restart after upgrading.
+
+- Add **官方工具协作** to the workbench launcher for **团队分工**. Choose read-only checks or approved editing in an isolated Git worktree. Preview and the session tool share `planOfficialTeam` / `executeOfficialTeamPlan`, filter CLI readiness for the selected mode, execute dependency packages sequentially and record results. This executor has no automatic model-directory API fallback; editable integration retains the existing clean-source and matching-baseline checks.
+- Preserve every vendor CLI's configured model by default. Catalog recommendation IDs are no longer passed as implicit CLI model arguments. Only explicit `cliModel`, saved route mappings or team `cliModelsJson` select a CLI model; package bindings take precedence over tool bindings and saved mappings. ZCode keeps its per-call model limitation.
+- Retain verified CLI model, entry, sandbox, usage and reported-cost evidence through routed, single-tool and team records. Separate recommended, requested and actual reported model identities. Preserve a valid reported zero cost; incomplete usage or unknown/unmatched actual CLI model identity cannot be costed using a recommended catalog price.
+- Bind installation/version and CLI login-status checks to the actual selected npm, pnpm or native copy. Refuse silent PATH substitution when the saved installation cannot be resolved or its concrete entry is unavailable.
+- Keep Windows Codex's original `CODEX_HOME` under the reviewed Harness ACL state-directory contract for read-only tasks. Explicitly selected native editable teams can use Codex's own workspace-write policy. Real Codex network checks failed in this batch; they are not reported as accepted execution.
+- Store Windows isolated Git worktrees in Host-derived per-source private temporary storage. Real use exposed source drives with Modify permission but no WRITE_OWNER, where the SDK could not initialize its ACL boundary. Keep source authorization and patch checks, reject redirected storage, and preserve failed worktrees for retry. A no-model SDK confined-write and cross-drive integration probe passed.
+- Add an explicit **厂商 CLI 正常环境** option for editable official teams. It retains normal vendor state, shell, skills and MCP under current user permissions, outside Harness's process sandbox. Git worktrees collect changes rather than restrict OS access. Startup and retry require separate confirmation even when legacy direct-CLI confirmation is disabled; restricted runs never switch automatically. Native Kimi and ordinary npm MiniMax entries additionally require registry package/manifest/entry attestation.
+- Preserve bounded, scrubbed failure diagnostics and exit codes, including Claude result errors. Keep verified usage and reported costs after failures, cancellation and timeouts, mark partial usage incomplete, and retain earlier team results if a later adapter throws. Reassignment applies the new route's saved CLI model mapping.
+- Clear obsolete failure details after a successful rerun and display the stored execution mode accurately. Update the English/Chinese README and workbench guide, including the existing Windows ZCode bundled TUI terminal support.
+
+Real Windows validation: authenticated workbench RPC executed one Claude package; the installed production runtime then executed Claude analysis → MiniMax delivery with dependency context and source patch integration. Independent JSON readback passed. Successful calls report $0.253848 known API cost in total; earlier unknown amounts and unpriced subscription reference usage remain separate. File/shell tools passed; every MCP service and vendor are not claimed tested. See [validation evidence and limits](docs/releases/0.20.0-validation.md). Automated checks and exact totals are included in the GitHub release notes.
+
+Engineering checks: frozen-lockfile strict-peer install passed; 833 tests, 830 passed, 3 conditional skips, no failures; client build and generated-client consistency passed. Skipped process/PTY checks are not counted as real execution acceptance.
+
 ## 0.19.0 — 2026-10-12
 
 Stable minor under the npm latest tag. Fully quit Harness (including its tray or WebUI Host process) and restart after upgrading.

package/README.i18n.yaml

@@ -1,3 +1,4 @@
 # README translation pairing for this repository.
-README.md: 4d820ba4402b344268124315ba74228dc6c2e79f
-README.zh.md: f00ed637e2f3a22b28cf7f5e4a4015f9c09eba0c
+# Reviewed together; Git blob hashes identify the exact synchronized texts.
+README.md: 4f6b76b9602764b9d3c8e360f17c873a5180e16b
+README.zh.md: 4e281f5632268114a86549380ba730df9ab69d07