wtt-connect 0.2.61 → 0.2.63

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "wtt-connect",
-  "version": "0.2.61",
+  "version": "0.2.63",
   "private": false,
   "description": "WTT-native connector daemon for Codex, Claude Code, Cursor, Gemini, ACP, and other coding agent surfaces.",
   "type": "module",

package/src/runner.js

@@ -21,6 +21,7 @@ import { runShellCommand } from './shell-runner.js';
 import { TerminalSessionManager } from './terminal-session.js';
 import { isModelSwitcherEnabled, switchModelProvider } from './model-switcher.js';
 import { formatSlashCommandHelp, isAgentSlashCommand, parseSlashCommand, slashCommandsForAdapter } from './slash.js';
+import { createUsageCollector, taskUsagePatch } from './usage.js';
 
 const TERMINAL_STATUSES = new Set(['review', 'done', 'approved', 'cancelled']);
 const GENERATED_FILE_EXTENSIONS = new Set([
@@ -82,7 +83,8 @@ export class Runner {
     if (!meta || meta.kb_mode !== true) return '';
     const kbTaskId = String(meta.kb_task_id || meta.kbTaskId || '').trim();
     if (!kbTaskId) return '';
-    const result = await this.api.searchKnowledgeContext(kbTaskId, query, { limit: 8 });
+    const explicitQuery = String(meta.kb_query || meta.kbQuery || '').trim();
+    const result = await this.api.searchKnowledgeContext(kbTaskId, explicitQuery || query, { limit: 8 });
     return renderKnowledgeContextBlock(result?.results || [], {
       taskId: kbTaskId,
       scope: meta.kb_scope || meta.kbScope || 'personal',
@@ -183,12 +185,14 @@ export class Runner {
     await this.wtt.typing(topicId, 'start', { statusText: 'Agent 已接收消息，正在准备执行', statusKind: 'queued', ttlMs: 30000 });
     let runtimeSelection = null;
     let chatStream = null;
+    let usageCollector = null;
     try {
       const transcripts = await this.transcribeAttachments(staged.files);
       const modelConfig = modelConfigFromMessage(m);
       const adapter = this.registry.select({ ...m, content, model_config: modelConfig });
+      usageCollector = createUsageCollector(adapter.name, modelConfig);
       runtimeSelection = { adapter: adapter.name, modelConfig };
-      chatStream = this.createChatStream(topicId, adapter.name, modelConfig);
+      chatStream = this.createChatStream(topicId, adapter.name, modelConfig, () => usageCollector?.summary());
       await chatStream.start();
       this.recordRuntimeSelection(adapter.name, modelConfig, 'running');
       await this.wtt.typing(topicId, 'start', { statusText: `${adapterDisplayName(adapter.name)} 正在执行${modelConfig.model ? ` · ${modelConfig.model}` : ''}`, statusKind: 'running', adapter: adapter.name, model: modelConfig.model || undefined, ttlMs: 30000 });
@@ -210,12 +214,16 @@ export class Runner {
           modelConfig,
           files: staged.files,
           images: staged.images,
-          onProgress: (event) => this.handleChatProgress(topicId, event, adapter.name, chatStream),
+          onProgress: (event) => {
+            usageCollector?.observe(event);
+            return this.handleChatProgress(topicId, event, adapter.name, chatStream);
+          },
         });
         if (localSlash !== null) {
           const reply = stripHiddenContextLeak(localSlash || '(empty response)') || '(empty response)';
           await chatStream.finish(reply);
           await this.wtt.publish(topicId, reply, 'CHAT_REPLY', chatStream.finalMetadata());
+          await this.reportUsageBestEffort(usageCollector?.summary(), { runId: chatStream.id, topicId, messageId: m.id });
           log('info', 'slash command replied', { topicId, adapter: adapter.name, chars: reply.length });
           return;
         }
@@ -254,13 +262,17 @@ export class Runner {
         modelConfig,
         files: staged.files,
         images: staged.images,
-        onProgress: (event) => this.handleChatProgress(topicId, event, adapter.name, chatStream),
+        onProgress: (event) => {
+          usageCollector?.observe(event);
+          return this.handleChatProgress(topicId, event, adapter.name, chatStream);
+        },
       });
       const prepared = prepareGeneratedFileArtifacts(stripHiddenContextLeak(output || '(empty response)') || '(empty response)');
       const reply = prepared.text || '(empty response)';
       await chatStream.finish(reply);
       await this.maybeAttachSpeech(topicId, reply, `chat-${topicId}`);
       await this.wtt.publish(topicId, reply, 'CHAT_REPLY', chatStream.finalMetadata());
+      await this.reportUsageBestEffort(usageCollector?.summary(), { runId: chatStream.id, topicId, messageId: m.id });
       await this.publishGeneratedFileArtifacts(topicId, prepared.refs, { source: 'chat', sourceId: topicId, adapter: adapter.name });
       log('info', 'chat replied', { topicId, chars: reply.length });
     } catch (err) {
@@ -283,6 +295,7 @@ export class Runner {
     const transcripts = await this.transcribeAttachments(staged.files);
     const modelConfig = modelConfigFromMessage(task);
     const adapter = this.registry.select({ ...task, model_config: modelConfig });
+    const usageCollector = createUsageCollector(adapter.name, modelConfig);
     this.recordRuntimeSelection(adapter.name, modelConfig, 'running');
     const agentProfile = await this.getAgentProfile();
     if (topicId) await this.wtt.typing(topicId, 'start', { statusText: `${adapterDisplayName(adapter.name)} 正在执行任务${modelConfig.model ? ` · ${modelConfig.model}` : ''}`, statusKind: 'running', adapter: adapter.name, model: modelConfig.model || undefined, ttlMs: 30000 });
@@ -313,14 +326,18 @@ export class Runner {
         modelConfig,
         files: staged.files,
         images: staged.images,
-        onProgress: (ev) => topicId ? this.maybePublishProgress(topicId, ev, adapter.name) : Promise.resolve(),
+        onProgress: (ev) => {
+          usageCollector.observe(ev);
+          return topicId ? this.maybePublishProgress(topicId, ev, adapter.name) : Promise.resolve();
+        },
       });
       const prepared = prepareGeneratedFileArtifacts(output || '(empty response)');
       const summary = prepared.text || '(empty response)';
       await this.maybeRelayArenaOpenCLResult(task, summary);
       const artifact = await this.materializeTaskArtifact(taskId, summary);
       const commitIds = extractCommitIds(summary);
-      const patch = { status: 'review', progress: 100, output: summary, summary, usage_source: `wtt-connect/${adapter.name}` };
+      const usage = usageCollector.summary();
+      const patch = { status: 'review', progress: 100, output: summary, summary, usage_source: `wtt-connect/${adapter.name}`, ...taskUsagePatch(usage) };
       if (commitIds.length) patch.commit_id = commitIds[0];
       await this.api.patchTask(taskId, patch);
       if (topicId) {
@@ -328,6 +345,7 @@ export class Runner {
         await this.publishGeneratedFileArtifacts(topicId, prepared.refs, { source: 'task', sourceId: taskId, adapter: adapter.name });
         if (artifact?.asset?.url) await this.wtt.publish(topicId, `Artifact: ${artifact.asset.url}`, 'TASK_ARTIFACT');
       }
+      await this.reportUsageBestEffort(usage, { runId: `task-${taskId}-${Date.now().toString(36)}`, topicId, taskId });
       log('info', 'task completed', { taskId, chars: summary.length });
     } catch (err) {
       await this.api.patchTask(taskId, { status: 'blocked', output: String(err.message || err) });
@@ -625,7 +643,7 @@ export class Runner {
     }
   }
 
-  createChatStream(topicId, adapterName = 'agent', modelConfig = {}) {
+  createChatStream(topicId, adapterName = 'agent', modelConfig = {}, usageProvider = null) {
     const enabled = this.config.messageStream !== false && Boolean(topicId);
     const streamId = `wtt-${Date.now().toString(36)}-${crypto.randomUUID().slice(0, 8)}`;
     let seq = 0;
@@ -648,11 +666,13 @@ export class Runner {
     return {
       id: streamId,
       finalMetadata() {
+        const usage = typeof usageProvider === 'function' ? usageProvider() : null;
         return {
           stream_id: streamId,
           streamId,
           adapter: adapterName,
           model: modelConfig.model || modelConfig.model_id || modelConfig.modelId || '',
+          ...(usage ? { usage } : {}),
         };
       },
       async start() {
@@ -698,6 +718,17 @@ export class Runner {
     ]);
   }
 
+  async reportUsageBestEffort(usage, { runId = '', topicId = '', taskId = '', messageId = '' } = {}) {
+    if (!usage) return;
+    await this.api.reportAgentUsage({
+      ...usage,
+      run_id: runId,
+      topic_id: topicId,
+      task_id: taskId,
+      message_id: messageId,
+    });
+  }
+
   async maybePublishProgress(topicId, event, adapterName = 'agent') {
     if (!topicId) return;
     const normalized = normalizeAgentEvent(adapterName, event);

package/src/usage.js

@@ -0,0 +1,207 @@
+export function createUsageCollector(adapter = 'agent', modelConfig = {}) {
+  let best = null;
+  let responseModel = '';
+  let rawUsage = null;
+
+  function observe(event) {
+    if (!event || typeof event !== 'object') return;
+    responseModel = extractModel(event) || responseModel;
+    const usage = extractUsage(event);
+    if (!usage) return;
+    rawUsage = usage.raw_usage || usage;
+    if (!best || scoreUsage(usage) >= scoreUsage(best)) best = usage;
+  }
+
+  function summary() {
+    if (!best) return null;
+    const inputTokens = safeInt(best.input_tokens);
+    const outputTokens = safeInt(best.output_tokens);
+    const cacheReadTokens = safeInt(best.cache_read_tokens);
+    const cacheWriteTokens = safeInt(best.cache_write_tokens);
+    const cacheTokens = safeInt(best.cache_tokens) + cacheReadTokens + cacheWriteTokens;
+    const totalTokens = safeInt(best.total_tokens) || inputTokens + outputTokens + cacheTokens;
+    if (!totalTokens && !inputTokens && !outputTokens && !cacheTokens) return null;
+    return {
+      adapter,
+      provider: providerForAdapter(adapter),
+      request_model: modelFromConfig(modelConfig),
+      response_model: responseModel,
+      input_tokens: inputTokens,
+      output_tokens: outputTokens,
+      cache_read_tokens: cacheReadTokens,
+      cache_write_tokens: cacheWriteTokens,
+      cache_tokens: cacheTokens,
+      total_tokens: totalTokens,
+      source: sourceForAdapter(adapter),
+      raw_usage: compactRawUsage(rawUsage),
+    };
+  }
+
+  return { observe, summary };
+}
+
+export function taskUsagePatch(usage) {
+  if (!usage) return {};
+  return {
+    usage_prompt_tokens: safeInt(usage.input_tokens),
+    usage_completion_tokens: safeInt(usage.output_tokens),
+    usage_cache_read_tokens: safeInt(usage.cache_read_tokens) + safeInt(usage.cache_tokens),
+    usage_cache_write_tokens: safeInt(usage.cache_write_tokens),
+    usage_total_tokens: safeInt(usage.total_tokens),
+    usage_source: usage.source || 'wtt-connect',
+  };
+}
+
+function extractUsage(event) {
+  const direct = findUsageObject(event);
+  if (direct) return normalizeUsage(direct);
+  const stats = event?.stats;
+  if (stats && typeof stats === 'object') {
+    const models = stats.models;
+    if (models && typeof models === 'object' && !Array.isArray(models)) {
+      return normalizeUsage(aggregateModelStats(models));
+    }
+    return normalizeUsage(stats);
+  }
+  return null;
+}
+
+function findUsageObject(value, depth = 0) {
+  if (!value || typeof value !== 'object' || depth > 5) return null;
+  for (const key of ['usage', 'usage_metadata', 'usageMetadata', 'token_usage', 'tokenUsage']) {
+    const candidate = value[key];
+    if (candidate && typeof candidate === 'object') return candidate;
+  }
+  if (looksLikeUsage(value)) return value;
+  for (const key of ['message', 'response', 'result', 'item', 'delta', 'event']) {
+    const found = findUsageObject(value[key], depth + 1);
+    if (found) return found;
+  }
+  return null;
+}
+
+function looksLikeUsage(value) {
+  if (!value || typeof value !== 'object') return false;
+  return [
+    'input_tokens',
+    'prompt_tokens',
+    'output_tokens',
+    'completion_tokens',
+    'total_tokens',
+    'promptTokenCount',
+    'candidatesTokenCount',
+    'totalTokenCount',
+  ].some((key) => key in value);
+}
+
+function normalizeUsage(raw) {
+  if (!raw || typeof raw !== 'object') return null;
+  const inputTokens = firstInt(raw, [
+    'input_tokens',
+    'prompt_tokens',
+    'promptTokens',
+    'prompt_token_count',
+    'promptTokenCount',
+    'inputTokenCount',
+  ]);
+  const candidateTokens = firstInt(raw, [
+    'output_tokens',
+    'completion_tokens',
+    'completionTokens',
+    'completion_token_count',
+    'candidatesTokenCount',
+    'outputTokenCount',
+  ]);
+  const thoughtsTokens = firstInt(raw, ['thoughtsTokenCount', 'reasoning_tokens', 'reasoningTokens']);
+  const outputTokens = candidateTokens + thoughtsTokens;
+  const cacheReadTokens = firstInt(raw, [
+    'cache_read_input_tokens',
+    'cached_tokens',
+    'cachedTokens',
+    'cachedContentTokenCount',
+  ]);
+  const cacheWriteTokens = firstInt(raw, ['cache_creation_input_tokens', 'cache_write_input_tokens']);
+  const cacheTokens = firstInt(raw, ['cache_tokens', 'cacheTokens']);
+  const totalTokens = firstInt(raw, ['total_tokens', 'totalTokens', 'totalTokenCount'])
+    || inputTokens + outputTokens + cacheReadTokens + cacheWriteTokens + cacheTokens;
+  if (!inputTokens && !outputTokens && !cacheReadTokens && !cacheWriteTokens && !cacheTokens && !totalTokens) return null;
+  return {
+    input_tokens: inputTokens,
+    output_tokens: outputTokens,
+    cache_read_tokens: cacheReadTokens,
+    cache_write_tokens: cacheWriteTokens,
+    cache_tokens: cacheTokens,
+    total_tokens: totalTokens,
+    raw_usage: raw,
+  };
+}
+
+function aggregateModelStats(models) {
+  const totals = {};
+  for (const value of Object.values(models || {})) {
+    if (!value || typeof value !== 'object') continue;
+    const usage = normalizeUsage(value);
+    if (!usage) continue;
+    for (const key of ['input_tokens', 'output_tokens', 'cache_read_tokens', 'cache_write_tokens', 'cache_tokens', 'total_tokens']) {
+      totals[key] = safeInt(totals[key]) + safeInt(usage[key]);
+    }
+  }
+  return totals;
+}
+
+function extractModel(event) {
+  if (!event || typeof event !== 'object') return '';
+  for (const key of ['model', 'model_id', 'modelId', 'current_model', 'response_model']) {
+    const value = event[key];
+    if (typeof value === 'string' && value.trim()) return value.trim();
+  }
+  return extractModel(event.message) || extractModel(event.response) || '';
+}
+
+function modelFromConfig(modelConfig = {}) {
+  return String(modelConfig.model || modelConfig.model_id || modelConfig.modelId || '').trim();
+}
+
+function firstInt(raw, keys) {
+  for (const key of keys) {
+    const value = raw?.[key];
+    const num = safeInt(value);
+    if (num) return num;
+  }
+  return 0;
+}
+
+function safeInt(value) {
+  const num = Number(value || 0);
+  if (!Number.isFinite(num) || num <= 0) return 0;
+  return Math.round(num);
+}
+
+function scoreUsage(usage) {
+  return safeInt(usage?.total_tokens) || safeInt(usage?.input_tokens) + safeInt(usage?.output_tokens) + safeInt(usage?.cache_tokens);
+}
+
+function providerForAdapter(adapter) {
+  if (adapter === 'claude-code') return 'anthropic-compatible';
+  if (adapter === 'codex') return 'openai-compatible';
+  if (adapter === 'gemini') return 'gemini';
+  return adapter || 'agent';
+}
+
+function sourceForAdapter(adapter) {
+  if (adapter === 'claude-code') return 'claude_code_stream_json';
+  if (adapter === 'codex') return 'codex_json';
+  if (adapter === 'gemini') return 'gemini_stream_json';
+  return `wtt-connect/${adapter || 'agent'}`;
+}
+
+function compactRawUsage(value) {
+  if (!value || typeof value !== 'object') return undefined;
+  try {
+    const json = JSON.stringify(value);
+    if (json.length <= 4000) return value;
+    return { truncated: true, total_tokens: safeInt(value.total_tokens || value.totalTokens || value.totalTokenCount) };
+  } catch {
+    return undefined;
+  }
+}

package/src/wtt-api.js

@@ -117,4 +117,17 @@ export class WTTApi {
     });
   }
 
+  async reportAgentUsage(usage) {
+    if (!usage || !this.config.agentId || !this.config.token) return null;
+    try {
+      return await this.request('POST', `/agents/${encodeURIComponent(this.config.agentId)}/usage`, {
+        headers: this.agentHeaders(),
+        json: usage,
+      });
+    } catch (err) {
+      log('warn', 'report_agent_usage failed', { error: err.message });
+      return null;
+    }
+  }
+
 }