@router-agent/infra 0.0.0-stage → 0.3.32

package/dist/vision.js

@@ -0,0 +1,164 @@
+/**
+ * 图片转文字（多模态模型）的基础设施。
+ *
+ * 只依赖 OpenAI 兼容的 `/chat/completions` 接口，不引入新 SDK：
+ * 飞书图片先由消息层下载成 base64，再交给这里转成可进入研判的文本。
+ */
+export const DEFAULT_VISION_PROMPT = [
+    '把这张聊天图片转换成可以直接进入研发研判的纯文本：',
+    '1. 逐字转写图片里所有可见文字，保留报错信息、日志、代码、命令行和表格结构；',
+    '2. 如果图片是界面截图，说明这是哪个页面/功能区域，以及图里表现出的异常或操作意图；',
+    '3. 如果图片是设计稿或流程图，按层级概述它的结构与关键信息。',
+    '看不清的内容写「[看不清]」，不要猜测或编造。',
+    '直接输出纯文本，不要加 Markdown 代码块，也不要解释你在做什么。',
+].join('\n');
+const DEFAULT_TIMEOUT_MS = 60_000;
+const DEFAULT_MAX_CHARS = 2_000;
+const DEFAULT_MAX_TOKENS = 1_500;
+/** 把 OpenAI 兼容的 baseUrl 规范成 chat/completions 地址。 */
+export function chatCompletionsUrl(baseUrl) {
+    const clean = baseUrl.trim().replace(/\/+$/, '');
+    if (!clean)
+        throw new Error('image:vision-base-url-missing');
+    if (clean.endsWith('/chat/completions'))
+        return clean;
+    return `${clean}/chat/completions`;
+}
+/** 从 `chat/completions` 的返回里取 assistant 文本；兼容字符串与分段数组。 */
+export function parseChatCompletionText(payload) {
+    const root = payload;
+    const content = root?.choices?.[0]?.message?.content;
+    if (typeof content === 'string')
+        return content.trim();
+    if (Array.isArray(content)) {
+        return content
+            .map((part) => (part && part.type === 'text' ? String(part.text ?? '') : ''))
+            .join('')
+            .trim();
+    }
+    return '';
+}
+/**
+ * 创建视觉转写器。
+ *
+ * 结构性失败（模型不支持图片、接口不存在）会熔断：同一进程里后续图片直接快速失败，
+ * 避免每张图都等一次超时、刷一屏同样的错误。
+ */
+export function createVisionTranscriber(model, opts = {}) {
+    const fetchImpl = opts.fetchImpl ?? globalThis.fetch;
+    const timeoutMs = opts.timeoutMs && opts.timeoutMs > 0 ? opts.timeoutMs : DEFAULT_TIMEOUT_MS;
+    const maxChars = opts.maxChars && opts.maxChars > 0 ? opts.maxChars : DEFAULT_MAX_CHARS;
+    const maxTokens = opts.maxTokens && opts.maxTokens > 0 ? opts.maxTokens : DEFAULT_MAX_TOKENS;
+    const prompt = opts.prompt?.trim() || DEFAULT_VISION_PROMPT;
+    const url = chatCompletionsUrl(model.baseUrl);
+    let unsupported = null;
+    // 一旦确认这个模型会把预算耗在 reasoning 上，后续图片直接关闭推理，省掉一次白跑的请求。
+    let reasoningOff = false;
+    /** 发一次请求；reasoningOff=true 时显式关闭推理模型的思考预算。 */
+    async function request(reasoningOff, image) {
+        const controller = new AbortController();
+        const timer = setTimeout(() => controller.abort(), timeoutMs);
+        let response;
+        try {
+            response = await fetchImpl(url, {
+                method: 'POST',
+                headers: {
+                    'Content-Type': 'application/json',
+                    ...(model.apiKey ? { Authorization: `Bearer ${model.apiKey}` } : {}),
+                },
+                body: JSON.stringify({
+                    model: model.model,
+                    temperature: model.temperature ?? 0.1,
+                    max_tokens: maxTokens,
+                    // 推理模型的思考 token 也算在 max_tokens 里；图片转写不需要推理，重试时直接关掉。
+                    ...(reasoningOff ? { reasoning_effort: 'none' } : {}),
+                    messages: [
+                        {
+                            role: 'user',
+                            content: [
+                                { type: 'text', text: prompt },
+                                { type: 'image_url', image_url: { url: `data:${image.mimeType};base64,${image.data}` } },
+                            ],
+                        },
+                    ],
+                }),
+                signal: controller.signal,
+            });
+        }
+        catch (error) {
+            if (isAbortError(error))
+                throw new Error(`image:vision-timeout(${timeoutMs}ms)`);
+            throw new Error(`image:vision-request-failed: ${errorText(error)}`);
+        }
+        finally {
+            clearTimeout(timer);
+        }
+        if (!response.ok) {
+            const detail = (await response.text().catch(() => '')).slice(0, 200);
+            // 404/415/422，或 400 且明确提到 image/vision，基本可判定为「这个模型/接口不接受图片」。
+            // 「图片太大」之类的 400 不熔断，换一张图仍可能成功。
+            const looksUnsupported = response.status === 404 || response.status === 415 || response.status === 422 ||
+                (response.status === 400 && /(not|doesn'?t|unsupported).{0,40}(support|image|vision)|modalit|multimodal/i.test(detail));
+            if (looksUnsupported)
+                unsupported = `http-${response.status}${detail ? `: ${detail}` : ''}`;
+            throw new Error(`image:vision-http-${response.status}${detail ? `: ${detail}` : ''}`);
+        }
+        try {
+            return await response.json();
+        }
+        catch (error) {
+            throw new Error(`image:vision-invalid-json: ${errorText(error)}`);
+        }
+    }
+    return {
+        async transcribe(image) {
+            if (unsupported)
+                throw new Error(`image:vision-unsupported: ${unsupported}`);
+            let payload = await request(reasoningOff, image);
+            let text = parseChatCompletionText(payload);
+            // 推理模型的 reasoning_content 会把 max_tokens 吃光，content 为空或半截就截断。
+            // 转写不需要思考，发现这种情况时明确关闭推理重试一次。
+            if (!reasoningOff && reasoningUsedUp(payload, text)) {
+                try {
+                    const retried = await request(true, image);
+                    const retriedText = parseChatCompletionText(retried);
+                    if (retriedText.length > text.length) {
+                        payload = retried;
+                        text = retriedText;
+                        reasoningOff = true;
+                    }
+                }
+                catch {
+                    // 部分 OpenAI 兼容服务不认识 reasoning_effort；重试失败保留首次结果。
+                }
+            }
+            if (!text)
+                throw new Error(`image:vision-empty(${completionDetail(payload)})`);
+            return text.length > maxChars ? `${text.slice(0, maxChars)}…` : text;
+        },
+    };
+}
+/** 推理 token 是否占满了预算：有 reasoning 消耗，且 content 为空或因为长度被截断。 */
+function reasoningUsedUp(payload, text) {
+    const root = payload;
+    const choice = root?.choices?.[0];
+    const finish = String(choice?.finish_reason ?? '');
+    const reasoning = Number(root?.usage?.completion_tokens_details?.reasoning_tokens ?? root?.usage?.reasoning_tokens ?? 0);
+    if (!Number.isFinite(reasoning) || reasoning <= 0)
+        return false;
+    return !text || finish === 'length';
+}
+/** 给空结果留一个可排查的细节：finish_reason 与 reasoning token 数。 */
+function completionDetail(payload) {
+    const root = payload;
+    const finish = String(root?.choices?.[0]?.finish_reason ?? 'unknown');
+    const reasoning = Number(root?.usage?.completion_tokens_details?.reasoning_tokens ?? root?.usage?.reasoning_tokens ?? 0);
+    return `finish=${finish}, reasoning=${Number.isFinite(reasoning) ? reasoning : 0}`;
+}
+function isAbortError(error) {
+    return error instanceof Error && (error.name === 'AbortError' || /aborted/i.test(error.message));
+}
+function errorText(error) {
+    return error instanceof Error ? error.message : String(error);
+}
+//# sourceMappingURL=vision.js.map

package/dist/vision.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"vision.js","sourceRoot":"","sources":["../../../../packages/infra/src/vision.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAiCH,MAAM,CAAC,MAAM,qBAAqB,GAAG;IACnC,2BAA2B;IAC3B,yCAAyC;IACzC,6CAA6C;IAC7C,iCAAiC;IACjC,yBAAyB;IACzB,sCAAsC;CACvC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AAEZ,MAAM,kBAAkB,GAAG,MAAM,CAAA;AACjC,MAAM,iBAAiB,GAAG,KAAK,CAAA;AAC/B,MAAM,kBAAkB,GAAG,KAAK,CAAA;AAEhC,oDAAoD;AACpD,MAAM,UAAU,kBAAkB,CAAC,OAAe;IAChD,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAA;IAChD,IAAI,CAAC,KAAK;QAAE,MAAM,IAAI,KAAK,CAAC,+BAA+B,CAAC,CAAA;IAC5D,IAAI,KAAK,CAAC,QAAQ,CAAC,mBAAmB,CAAC;QAAE,OAAO,KAAK,CAAA;IACrD,OAAO,GAAG,KAAK,mBAAmB,CAAA;AACpC,CAAC;AAED,0DAA0D;AAC1D,MAAM,UAAU,uBAAuB,CAAC,OAAgB;IACtD,MAAM,IAAI,GAAG,OAAqC,CAAA;IAClD,MAAM,OAAO,GAAG,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAA;IACpD,IAAI,OAAO,OAAO,KAAK,QAAQ;QAAE,OAAO,OAAO,CAAC,IAAI,EAAE,CAAA;IACtD,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QAC3B,OAAO,OAAO;aACX,GAAG,CAAC,CAAC,IAA6B,EAAE,EAAE,CAAC,CAAC,IAAI,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;aACrG,IAAI,CAAC,EAAE,CAAC;aACR,IAAI,EAAE,CAAA;IACX,CAAC;IACD,OAAO,EAAE,CAAA;AACX,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,uBAAuB,CAAC,KAAwB,EAAE,OAAsB,EAAE;IACxF,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,UAAU,CAAC,KAAK,CAAA;IACpD,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAA;IAC5F,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,iBAAiB,CAAA;IACvF,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,kBAAkB,CAAA;IAC5F,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,IAAI,EAAE,IAAI,qBAAqB,CAAA;IAC3D,MAAM,GAAG,GAAG,kBAAkB,CAAC,KAAK,CAAC,OAAO,CAAC,CAAA;IAC7C,IAAI,WAAW,GAAkB,IAAI,CAAA;IACrC,mDAAmD;IACnD,IAAI,YAAY,GAAG,KAAK,CAAA;IAExB,8CAA8C;IAC9C,KAAK,UAAU,OAAO,CAAC,YAAqB,EAAE,KAAiB;QAC7D,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAA;QACxC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAA;QAC7D,IAAI,QAAkB,CAAA;QACtB,IAAI,CAAC;YACH,QAAQ,GAAG,MAAM,SAAS,CAAC,GAAG,EAAE;gBAC9B,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE;oBACP,cAAc,EAAE,kBAAkB;oBAClC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,aAAa,EAAE,UAAU,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;iBACrE;gBACD,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC;oBACnB,KAAK,EAAE,KAAK,CAAC,KAAK;oBAClB,WAAW,EAAE,KAAK,CAAC,WAAW,IAAI,GAAG;oBACrC,UAAU,EAAE,SAAS;oBACrB,oDAAoD;oBACpD,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;oBACrD,QAAQ,EAAE;wBACR;4BACE,IAAI,EAAE,MAAM;4BACZ,OAAO,EAAE;gCACP,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE;gCAC9B,EAAE,IAAI,EAAE,WAAW,EAAE,SAAS,EAAE,EAAE,GAAG,EAAE,QAAQ,KAAK,CAAC,QAAQ,WAAW,KAAK,CAAC,IAAI,EAAE,EAAE,EAAE;6BACzF;yBACF;qBACF;iBACF,CAAC;gBACF,MAAM,EAAE,UAAU,CAAC,MAAM;aAC1B,CAAC,CAAA;QACJ,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,IAAI,YAAY,CAAC,KAAK,CAAC;gBAAE,MAAM,IAAI,KAAK,CAAC,wBAAwB,SAAS,KAAK,CAAC,CAAA;YAChF,MAAM,IAAI,KAAK,CAAC,gCAAgC,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC,CAAA;QACrE,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAA;QACrB,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;YACjB,MAAM,MAAM,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAA;YACpE,6DAA6D;YAC7D,+BAA+B;YAC/B,MAAM,gBAAgB,GAAG,QAAQ,CAAC,MAAM,KAAK,GAAG,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG;gBACpG,CAAC,QAAQ,CAAC,MAAM,KAAK,GAAG,IAAI,6EAA6E,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAA;YACzH,IAAI,gBAAgB;gBAAE,WAAW,GAAG,QAAQ,QAAQ,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAA;YAC3F,MAAM,IAAI,KAAK,CAAC,qBAAqB,QAAQ,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAA;QACvF,CAAC;QAED,IAAI,CAAC;YACH,OAAO,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAA;QAC9B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,IAAI,KAAK,CAAC,8BAA8B,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC,CAAA;QACnE,CAAC;IACH,CAAC;IAED,OAAO;QACL,KAAK,CAAC,UAAU,CAAC,KAAiB;YAChC,IAAI,WAAW;gBAAE,MAAM,IAAI,KAAK,CAAC,6BAA6B,WAAW,EAAE,CAAC,CAAA;YAE5E,IAAI,OAAO,GAAG,MAAM,OAAO,CAAC,YAAY,EAAE,KAAK,CAAC,CAAA;YAChD,IAAI,IAAI,GAAG,uBAAuB,CAAC,OAAO,CAAC,CAAA;YAC3C,6DAA6D;YAC7D,6BAA6B;YAC7B,IAAI,CAAC,YAAY,IAAI,eAAe,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,CAAC;gBACpD,IAAI,CAAC;oBACH,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,IAAI,EAAE,KAAK,CAAC,CAAA;oBAC1C,MAAM,WAAW,GAAG,uBAAuB,CAAC,OAAO,CAAC,CAAA;oBACpD,IAAI,WAAW,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC;wBACrC,OAAO,GAAG,OAAO,CAAA;wBACjB,IAAI,GAAG,WAAW,CAAA;wBAClB,YAAY,GAAG,IAAI,CAAA;oBACrB,CAAC;gBACH,CAAC;gBAAC,MAAM,CAAC;oBACP,iDAAiD;gBACnD,CAAC;YACH,CAAC;YACD,IAAI,CAAC,IAAI;gBAAE,MAAM,IAAI,KAAK,CAAC,sBAAsB,gBAAgB,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;YAC9E,OAAO,IAAI,CAAC,MAAM,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAA;QACtE,CAAC;KACF,CAAA;AACH,CAAC;AAED,4DAA4D;AAC5D,SAAS,eAAe,CAAC,OAAgB,EAAE,IAAY;IACrD,MAAM,IAAI,GAAG,OAAiD,CAAA;IAC9D,MAAM,MAAM,GAAG,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,CAAA;IACjC,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,aAAa,IAAI,EAAE,CAAC,CAAA;IAClD,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,yBAAyB,EAAE,gBAAgB,IAAI,IAAI,EAAE,KAAK,EAAE,gBAAgB,IAAI,CAAC,CAAC,CAAA;IACxH,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,SAAS,IAAI,CAAC;QAAE,OAAO,KAAK,CAAA;IAC/D,OAAO,CAAC,IAAI,IAAI,MAAM,KAAK,QAAQ,CAAA;AACrC,CAAC;AAED,uDAAuD;AACvD,SAAS,gBAAgB,CAAC,OAAgB;IACxC,MAAM,IAAI,GAAG,OAAiD,CAAA;IAC9D,MAAM,MAAM,GAAG,MAAM,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,aAAa,IAAI,SAAS,CAAC,CAAA;IACrE,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,yBAAyB,EAAE,gBAAgB,IAAI,IAAI,EAAE,KAAK,EAAE,gBAAgB,IAAI,CAAC,CAAC,CAAA;IACxH,OAAO,UAAU,MAAM,eAAe,MAAM,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,CAAA;AACpF,CAAC;AAED,SAAS,YAAY,CAAC,KAAc;IAClC,OAAO,KAAK,YAAY,KAAK,IAAI,CAAC,KAAK,CAAC,IAAI,KAAK,YAAY,IAAI,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAA;AAClG,CAAC;AAED,SAAS,SAAS,CAAC,KAAc;IAC/B,OAAO,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAA;AAC/D,CAAC"}

package/package.json

@@ -1,6 +1,33 @@
 {
   "name": "@router-agent/infra",
-  "version": "0.0.0-stage",
-  "stub": true,
-  "description": "Temporary package placeholder for staged publishing"
-}
+  "version": "0.3.32",
+  "type": "module",
+  "description": "基础设施：日志、子进程、对话模型与图片识别",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/blankPen/router-agent.git",
+    "directory": "packages/infra"
+  },
+  "homepage": "https://github.com/blankPen/router-agent#readme",
+  "bugs": {
+    "url": "https://github.com/blankPen/router-agent/issues"
+  },
+  "main": "./dist/index.js",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js"
+    },
+    "./*.js": "./dist/*.js",
+    "./*": "./dist/*"
+  },
+  "files": [
+    "dist"
+  ],
+  "types": "./dist/index.d.ts",
+  "publishConfig": {
+    "access": "public",
+    "registry": "https://registry.npmjs.org/"
+  }
+}

package/README.md

@@ -1,3 +0,0 @@
-# Temporary Holding Version
-
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.