@trim21/personal-pi-extensions 0.0.173 → 0.0.175

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@trim21/personal-pi-extensions",
-  "version": "0.0.173",
+  "version": "0.0.175",
   "type": "module",
   "description": "Custom pi coding-agent extensions: bwrap sandbox, workspace guard, opencode edit, and more",
   "keywords": [

package/src/vision-agent.ts

@@ -223,6 +223,39 @@ export function isMultimodal(model: { input?: readonly string[] } | undefined):
   return !!model && Array.isArray(model.input) && model.input.includes("image");
 }
 
+/**
+ * 把识别结果整理成 pendant 渲染用的 markdown。
+ * description 格式为 callVision 拼装的 `[label]\n<正文>\n[模型: ...]`，
+ * 第一行是图片标签、最后一行是模型/ token 元信息，中间是视觉模型输出正文。
+ */
+export function buildPendantMarkdown(params: {
+  paths: string[];
+  prompt: string;
+  description: string;
+  provider: string;
+  model: string;
+}): string {
+  const lines = params.description.split("\n");
+  const labels = lines[0]?.trim().replaceAll(/^\[|\]$/g, "") ?? "";
+  const footer = lines.at(-1)?.trim() ?? "";
+  const body = lines.slice(1, -1).join("\n").trim();
+  const files = params.paths.map((p) => `\`${basename(p)}\``).join(", ");
+  return [
+    "## 图片识别",
+    "",
+    `**图片**: ${files || labels}`,
+    `**视觉模型**: \`${params.provider}/${params.model}\``,
+    `**Prompt**: ${params.prompt}`,
+    "",
+    "### 识别结果",
+    "",
+    body || params.description,
+    footer ? `\n---\n${footer}` : "",
+  ]
+    .join("\n")
+    .trim();
+}
+
 // ── 图片加载与 API 调用 ─────────────────────────────────────────────────────
 
 /** 按文件头识别真实图片类型，识别不出返回 undefined */
@@ -457,7 +490,7 @@ export default function visionAgent(pi: ExtensionAPI) {
       ),
     }),
 
-    async execute(_toolCallId, params, signal, _onUpdate, ctx) {
+    async execute(_toolCallId, params, signal, onUpdate, ctx) {
       try {
         const paths = resolveImagePaths({ path: params.path });
         if (paths.length === 0) {
@@ -502,19 +535,39 @@ export default function visionAgent(pi: ExtensionAPI) {
           };
         }
 
+        const prompt = buildPrompt(params.prompt, paths.length);
+        // 把发给视觉模型的 prompt 实时透传给主会话，识别要求对主模型可见
+        onUpdate?.({
+          content: [{ type: "text", text: `视觉识别 prompt: ${prompt}` }],
+          details: { prompt },
+        });
+
         const description = await callVision(
           { ...provider, model: visionConfig.model },
           paths,
-          buildPrompt(params.prompt, paths.length),
+          prompt,
           signal ?? ctx.signal,
         );
+        const markdown = buildPendantMarkdown({
+          paths,
+          prompt,
+          description,
+          provider: providerName,
+          model: visionConfig.model,
+        });
         return {
           content: [{ type: "text", text: description }],
           details: {
             provider: providerName,
             model: visionConfig.model,
+            prompt,
+            output: description,
             paths,
             count: paths.length,
+            pendant: {
+              markdown,
+              expanded: true,
+            },
           },
         };
       } catch (error) {
@@ -529,7 +582,13 @@ export default function visionAgent(pi: ExtensionAPI) {
         return {
           isError: true,
           content: [{ type: "text", text: `识别失败: ${message}` }],
-          details: { error: message },
+          details: {
+            error: message,
+            pendant: {
+              markdown: `## 图片识别失败\n\n${message}`,
+              expanded: true,
+            },
+          },
         };
       }
     },