@oh-my-pi/pi-coding-agent 17.3.2 → 17.3.4

package/dist/types/markit/converters/pdf/grid.d.ts

@@ -1,25 +0,0 @@
-/**
- * Table grid detection from vector segments and text boxes.
- *
- * Ported from @oharato/pdf2md-ts with TypeScript types and without
- * CJK-specific borderless table heuristics. The core algorithm:
- *
- * 1. Classify segments as horizontal or vertical lines
- * 2. Group horizontal Y-lines into table groups (split by vertical gaps)
- * 3. For each group:
- *    a. Full grid (H+V lines): build cells from grid intersections,
- *       place text via raycasting
- *    b. H-line only (no V lines): infer columns from text X positions
- * 4. Prune empty rows/cols
- *
- * Coordinate system: PDF native (bottom-left origin, Y increases upward).
- */
-import type { Segment, TableGrid, TextBox } from "./types.js";
-export interface GridResult {
-    grids: TableGrid[];
-    consumedIds: string[];
-}
-/**
- * Detect all table grids on a single page from its text boxes and segments.
- */
-export declare function resolveTableGrids(pageNumber: number, textBoxes: TextBox[], segments: Segment[]): GridResult;

package/dist/types/markit/converters/pdf/headers.d.ts

@@ -1,24 +0,0 @@
-/**
- * Running header/footer detection and removal.
- *
- * Many PDFs have repeated text at the top or bottom of every page:
- * document titles, chapter names, page numbers, copyright notices.
- * These pollute the markdown output as false headings or noise.
- *
- * Algorithm:
- *   1. For each page, bucket text boxes by Y position (top/bottom zones)
- *   2. Collect the text content at each zone across all pages
- *   3. Text appearing on >20% of pages OR 8+ consecutive pages is a
- *      running header/footer
- *   4. Remove matching text boxes before further processing
- */
-import type { PageContent } from "./types.js";
-/**
- * Detect and remove running headers and footers from all pages.
- * Mutates the pages array in place, removing header/footer text boxes.
- *
- * Uses two strategies:
- *   1. Global frequency: text appearing on > 20% of all pages
- *   2. Consecutive runs: text appearing on 8+ consecutive pages
- */
-export declare function stripHeadersFooters(pages: PageContent[]): void;

package/dist/types/markit/converters/pdf/render.d.ts

@@ -1,24 +0,0 @@
-/**
- * Markdown rendering for PDF pages.
- *
- * Converts table grids and free text boxes into markdown, handling:
- * - Table grid → markdown table (`| col | col |`)
- * - Free text → paragraphs with heading detection (by font size)
- * - Content ordering (top-to-bottom via Y coordinate)
- * - Paragraph wrap merging (lines broken across PDF line boundaries)
- * - Page number removal
- *
- * Ported from @oharato/pdf2md-ts, stripped of CJK/TDnet-specific logic.
- */
-import type { TableGrid, TextBox } from "./types.js";
-/**
- * Render a TableGrid as a markdown table.
- */
-export declare function renderTableToMarkdown(table: TableGrid): string;
-/**
- * Render one page's content: free text and tables interleaved top-to-bottom.
- */
-export declare function renderPageContent(freeTextBoxes: TextBox[], tables: TableGrid[], imageBlocks?: Array<{
-    topY: number;
-    markdown: string;
-}>, allTextBoxes?: TextBox[]): string;

package/dist/types/markit/converters/pdf/types.d.ts

@@ -1,75 +0,0 @@
-/** Bounding box in PDF coordinate space (origin = bottom-left). */
-export type Bounds = {
-    left: number;
-    right: number;
-    /** Higher value = higher on the page. */
-    top: number;
-    bottom: number;
-};
-/** A text fragment with position and font metadata. */
-export type TextBox = {
-    id: string;
-    text: string;
-    bounds: Bounds;
-    pageNumber: number;
-    /** Dominant font size in points. */
-    fontSize: number;
-    /** True if rendered bold (font name or rendering mode). */
-    isBold: boolean;
-};
-/** A horizontal or vertical line segment extracted from vector graphics. */
-export type Segment = {
-    id: string;
-    x1: number;
-    y1: number;
-    x2: number;
-    y2: number;
-};
-/** A single cell in a resolved table grid. */
-export type TableCell = {
-    row: number;
-    col: number;
-    text: string;
-    rowSpan: number;
-    colSpan: number;
-};
-/** A resolved table grid ready for markdown rendering. */
-export type TableGrid = {
-    pageNumber: number;
-    rows: number;
-    cols: number;
-    cells: TableCell[];
-    warnings: string[];
-    /** Top Y coordinate (PDF space: larger = higher on page). */
-    topY: number;
-    /** True for tables detected without vector borders. */
-    isBorderless: boolean;
-};
-/** An image/diagram region detected on a page. */
-export type ImageRegion = {
-    id: string;
-    pageNumber: number;
-    /** Bounding box in mupdf coordinates (top-left origin). */
-    bbox: {
-        x: number;
-        y: number;
-        w: number;
-        h: number;
-    };
-    /** Y position in PDF coordinates (bottom-left) for ordering. */
-    topY: number;
-};
-/** Result of extracting content from a single PDF page. */
-export type PageContent = {
-    pageNumber: number;
-    textBoxes: TextBox[];
-    segments: Segment[];
-    images: ImageRegion[];
-};
-/** A block of rendered content (text paragraph or table). */
-export type ContentBlock = {
-    topY: number;
-    content: string;
-    /** True if this line has wide gaps between text boxes (column headers). */
-    isTabular?: boolean;
-};

package/dist/types/tools/read-pdf-images.d.ts

@@ -1,12 +0,0 @@
-import type { AgentToolResult } from "@oh-my-pi/pi-agent-core";
-import type { ToolSession } from "../sdk.js";
-import type { ReadToolDetails } from "./read.js";
-export declare function rewritePdfImagePlaceholders(markdown: string, pdfPath: string): string;
-export declare function splitPdfImageMemberReadPath(readPath: string): {
-    pdfPath: string;
-    member: string;
-} | null;
-export declare function readPdfImageMember(session: ToolSession, autoResizeImages: boolean, absolutePdfPath: string, pdfDisplayPath: string, member: string, suffixResolution: {
-    from: string;
-    to: string;
-} | undefined, signal?: AbortSignal): Promise<AgentToolResult<ReadToolDetails>>;

package/dist/types/utils/mupdf-wasm-embed.d.ts

@@ -1 +0,0 @@
-export declare function loadEmbeddedMupdfWasm(): Uint8Array | undefined;

package/scripts/embed-mupdf-wasm.ts

@@ -1,67 +0,0 @@
-#!/usr/bin/env bun
-
-// Embeds mupdf's `mupdf-wasm.wasm` into the compiled single-file binary.
-//
-// mupdf loads its wasm by reading the `mupdf-wasm.wasm` sibling of its own
-// module via `new URL(..., import.meta.url)` + `readFileSync`. A `bun --compile`
-// binary has no node_modules, so that read fails (`ENOENT .../mupdf-wasm.wasm`),
-// and marking mupdf `--external` instead makes `bun --compile` eagerly fail to
-// resolve the package at startup (the static `import * as mupdf` lives in a lazy
-// chunk but is hoisted). So the binary build bundles mupdf and embeds the wasm
-// bytes here, handing them to the WASM module as `$libmupdf_wasm_Module.wasmBinary`
-// (see src/utils/markit.ts).
-//
-// `--generate` copies the wasm next to src/utils/mupdf-wasm-embed.ts and rewrites
-// that module to import it via `with { type: "file" }`; `--reset` restores the
-// checked-in placeholder and removes the copy. The npm `dist/cli.js` bundle never
-// runs this — it keeps mupdf external and loads the wasm from node_modules.
-
-import * as fs from "node:fs/promises";
-import { createRequire } from "node:module";
-import * as path from "node:path";
-
-const utilsDir = path.join(import.meta.dir, "..", "src", "utils");
-const helperPath = path.join(utilsDir, "mupdf-wasm-embed.ts");
-const wasmCopyPath = path.join(utilsDir, "mupdf-wasm.wasm");
-
-const placeholder = `// AUTOGENERATED -- managed by scripts/embed-mupdf-wasm.ts. Do not edit by hand.
-//
-// Compiled single-file binaries cannot let mupdf resolve its \`mupdf-wasm.wasm\`
-// sibling from the read-only bunfs, so the binary build (scripts/build-binary.ts
-// and scripts/ci-release-build-binaries.ts) regenerates this module to embed the
-// wasm bytes via \`with { type: "file" }\` and copies the wasm next to it. Source
-// checkouts, \`bun test\`, and the npm \`dist/cli.js\` bundle keep mupdf external and
-// load the wasm from node_modules, so this placeholder returns undefined and the
-// build resets back to it afterward.
-export function loadEmbeddedMupdfWasm(): Uint8Array | undefined {
-\treturn undefined;
-}
-`;
-
-const generated = `// AUTOGENERATED -- managed by scripts/embed-mupdf-wasm.ts. Do not edit or commit.
-import { readFileSync } from "node:fs";
-import wasmPath from "./mupdf-wasm.wasm" with { type: "file" };
-
-export function loadEmbeddedMupdfWasm(): Uint8Array | undefined {
-\treturn readFileSync(wasmPath);
-}
-`;
-
-if (process.argv.includes("--reset")) {
-	await Bun.write(helperPath, placeholder);
-	try {
-		await fs.unlink(wasmCopyPath);
-	} catch (err) {
-		if ((err as NodeJS.ErrnoException).code !== "ENOENT") throw err;
-	}
-	process.exit(0);
-}
-
-const wasmSource = path.join(path.dirname(createRequire(import.meta.url).resolve("mupdf")), "mupdf-wasm.wasm");
-const wasmFile = Bun.file(wasmSource);
-if (!(await wasmFile.exists())) {
-	throw new Error(`mupdf wasm not found at ${wasmSource}; run \`bun install\` first.`);
-}
-await Bun.write(wasmCopyPath, wasmFile);
-await Bun.write(helperPath, generated);
-console.log(`Embedded mupdf wasm (${wasmFile.size} bytes) into ${path.relative(process.cwd(), wasmCopyPath)}`);

package/src/markit/converters/pdf/columns.ts

@@ -1,103 +0,0 @@
-// Adapted from markit-ai (MIT). See ../../NOTICE.
-
-/**
- * Multi-column layout detection and text box reordering.
- *
- * Many PDFs (legal documents, datasheets, academic papers) use two-column
- * layouts. Without column detection, text boxes are ordered by Y position
- * only, interleaving left and right column content.
- *
- * Algorithm:
- *   1. Collect left edges of all text boxes on the page
- *   2. Find the largest horizontal gap between consecutive left edges
- *   3. If gap > MIN_GAP_RATIO of the text width and both sides have
- *      enough boxes → multi-column detected
- *   4. Assign each text box to a column based on its center X
- *   5. Return columns in reading order (left-to-right, top-to-bottom)
- *
- * This only detects the column structure. The caller is responsible for
- * processing each column's text boxes independently (table detection,
- * rendering, etc.).
- */
-import type { TextBox } from "./types";
-
-export interface ColumnLayout {
-	/** Number of columns detected (1 = single column, 2+ = multi-column). */
-	columnCount: number;
-	/** Text boxes grouped by column, in reading order (left to right). */
-	columns: TextBox[][];
-	/** X positions of column boundaries (between columns). */
-	boundaries: number[];
-}
-
-/**
- * Minimum gap as a fraction of the total text width to consider a column
- * boundary. A two-column layout typically has ~50% gap; we use a lower
- * threshold to catch asymmetric columns.
- */
-const MIN_GAP_RATIO = 0.15;
-/** Minimum number of text boxes on each side of the gap. */
-const MIN_BOXES_PER_COLUMN = 4;
-/** Minimum gap in absolute points to avoid splitting on small whitespace. */
-const MIN_GAP_PTS = 40;
-
-/**
- * Detect column layout and return text boxes grouped by column.
- *
- * For single-column pages, returns all boxes in one group.
- * For multi-column pages, returns boxes split by column in reading order.
- */
-export function detectColumns(textBoxes: TextBox[]): ColumnLayout {
-	if (textBoxes.length < MIN_BOXES_PER_COLUMN * 2) {
-		return { columnCount: 1, columns: [textBoxes], boundaries: [] };
-	}
-	// Collect unique left edges (rounded to avoid float noise)
-	const lefts = [...new Set(textBoxes.map(tb => Math.round(tb.bounds.left)))].sort((a, b) => a - b);
-	if (lefts.length < 2) {
-		return { columnCount: 1, columns: [textBoxes], boundaries: [] };
-	}
-	const textXMin = lefts[0];
-	const textXMax = Math.max(...textBoxes.map(tb => Math.round(tb.bounds.right)));
-	const textWidth = textXMax - textXMin;
-	if (textWidth <= 0) {
-		return { columnCount: 1, columns: [textBoxes], boundaries: [] };
-	}
-	// Find the largest gap between consecutive left-edge positions
-	let maxGap = 0;
-	let gapLeft = 0;
-	let gapRight = 0;
-	for (let i = 1; i < lefts.length; i++) {
-		const gap = lefts[i] - lefts[i - 1];
-		if (gap > maxGap) {
-			maxGap = gap;
-			gapLeft = lefts[i - 1];
-			gapRight = lefts[i];
-		}
-	}
-	const gapRatio = maxGap / textWidth;
-	if (gapRatio < MIN_GAP_RATIO || maxGap < MIN_GAP_PTS) {
-		return { columnCount: 1, columns: [textBoxes], boundaries: [] };
-	}
-	// Split point is the midpoint of the gap
-	const splitX = (gapLeft + gapRight) / 2;
-	// Assign boxes to columns based on center X
-	const leftCol: TextBox[] = [];
-	const rightCol: TextBox[] = [];
-	for (const tb of textBoxes) {
-		const cx = (tb.bounds.left + tb.bounds.right) / 2;
-		if (cx < splitX) {
-			leftCol.push(tb);
-		} else {
-			rightCol.push(tb);
-		}
-	}
-	// Validate both columns have enough content
-	if (leftCol.length < MIN_BOXES_PER_COLUMN || rightCol.length < MIN_BOXES_PER_COLUMN) {
-		return { columnCount: 1, columns: [textBoxes], boundaries: [] };
-	}
-	return {
-		columnCount: 2,
-		columns: [leftCol, rightCol],
-		boundaries: [splitX],
-	};
-}