"use strict";

import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
import { Type } from "typebox";
import { getConfig } from "../config";
import { processImage } from "../media";
import { visionCompletion } from "../client";
import { GENERAL_IMAGE_ANALYSIS } from "../prompts";
import {
  DEFAULT_MAX_BYTES,
  DEFAULT_MAX_LINES,
  truncateHead,
} from "@earendil-works/pi-coding-agent";

export function registerImageAnalysisTool(pi: ExtensionAPI) {
  pi.registerTool({
    name: "image_analysis",
    label: "Image Analysis",
    description:
      "Analyze any image and provide detailed description, object identification, context understanding, or answer specific questions about visual content. Supports local files (png, jpg) and HTTP(S) URLs.",
    promptSnippet: "Analyze images with GLM-4.6V vision — describe/identify/extract/answer based on user prompt",
    promptGuidelines: [
      "Use image_analysis when the user asks you to describe an image, identify objects, or answer questions about visual content. This is the general-purpose vision tool — use it when no more specific vision tool (extract_text_from_screenshot, diagnose_error_screenshot, etc.) fits the task.",
    ],
    parameters: Type.Object({
      imagePath: Type.String({
        description:
          "Path to the image file (relative to cwd, supports @ prefix) or an HTTP(S) URL.",
      }),
      prompt: Type.String({
        description:
          "What should the AI analyze or answer about this image? Be specific about what you want to know.",
      }),
    }),
    async execute(_toolCallId, params, signal, onUpdate, ctx) {
      const config = getConfig();

      onUpdate?.({
        content: [{ type: "text", text: "正在调用 GLM-4.6V 分析图片..." }],
      });

      const image = processImage(params.imagePath, ctx.cwd, config.maxImageSizeMB);

      const result = await visionCompletion(
        config,
        GENERAL_IMAGE_ANALYSIS,
        [image],
        params.prompt,
        signal
      );

      const truncated = truncateHead(result, {
        maxLines: DEFAULT_MAX_LINES,
        maxBytes: DEFAULT_MAX_BYTES,
      });

      let text = truncated.content;
      if (truncated.truncated) {
        text += `\n\n[输出已截断: ${truncated.outputLines}/${truncated.totalLines} 行]`;
      }

      return {
        content: [{ type: "text", text }],
        details: { raw: result },
      };
    },
  });
}