"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 { TEXT_EXTRACTION } from "../prompts";
import {
  DEFAULT_MAX_BYTES,
  DEFAULT_MAX_LINES,
  truncateHead,
} from "@earendil-works/pi-coding-agent";

export function registerExtractTextTool(pi: ExtensionAPI) {
  pi.registerTool({
    name: "extract_text_from_screenshot",
    label: "Extract Text",
    description:
      "Extract and transcribe all visible text from a screenshot with maximum accuracy. Preserves original formatting, indentation, and structure. Ideal for code extraction, terminal output, config files, and documentation screenshots. Supports local files (png, jpg) and HTTP(S) URLs.",
    promptSnippet: "Extract text/code from screenshots with OCR — preserves indentation and structure",
    promptGuidelines: [
      "Use extract_text_from_screenshot when the user provides a screenshot containing code, terminal output, logs, config files, or documentation that needs to be transcribed as text. This tool also accepts an optional programmingLanguage hint for better code extraction accuracy.",
    ],
    parameters: Type.Object({
      imagePath: Type.String({
        description:
          "Path to the screenshot (relative to cwd, supports @ prefix) or HTTP(S) URL.",
      }),
      prompt: Type.Optional(
        Type.String({
          description:
            "Specific extraction instructions. If omitted, defaults to 'Extract all visible text from this screenshot.'",
        })
      ),
      programmingLanguage: Type.Optional(
        Type.String({
          description:
            "Programming language hint for better code extraction accuracy (e.g., 'python', 'rust', 'typescript').",
        })
      ),
    }),
    async execute(_toolCallId, params, signal, onUpdate, ctx) {
      const config = getConfig();

      onUpdate?.({
        content: [{ type: "text", text: "正在调用 GLM-4.6V 提取文字..." }],
      });

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

      // 构建 user prompt
      let userPrompt = params.prompt || "Extract all visible text from this screenshot.";
      if (params.programmingLanguage) {
        userPrompt = `Programming language context: ${params.programmingLanguage}\n\n${userPrompt}`;
      }

      const result = await visionCompletion(
        config,
        TEXT_EXTRACTION,
        [image],
        userPrompt,
        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 },
      };
    },
  });
}