* Parse MULTIFORMATS GDI comment — contains format descriptors pointing to embedded data.

\o(base, \s\up N(overlay)) → overrightarrow etc.

* Parse a single text paragraph (`a:p`).

* Parse placeholder info from nvPr > p:ph.

\r(degree, radicand) or \r(,radicand) → \sqrt[degree]{radicand}

Parse one MODEL_ROUTES value (string model, or {model, thinking}) into a StageRoute.

* Parse a table row (`a:tr`).

\s\up N(text) → ^{text} or \s\do N(text) → _{text}

* Parse Shiki HTML output into per-line HTML fragments. * Shiki outputs: <pre ...><code><span class="line">...</span>\n...</code></pre> * We split b

* Parse a STRETCHDIBITS record and extract the bitmap as RGBA buffer.

* Break cell text into lines based upon table column width (e.g.: Magic Happens Here(tm)) * @param {TableCell} cell - table cell * @param {number} c

Validate/sanitize a route's `thinking` object into a ThinkingConfig (drops bad fields with a warn).

* Parse PDF using self-hosted MinerU service (mineru-api) * * Official MinerU API endpoint: * POST /file_parse (multipart/form-data) * * Respons

* Parse PDF using unpdf (existing implementation)

* Bounding box of a path's coordinates (path/viewBox space). Uses control * points for curves (a safe superset). Handles the commands our shapes emit

* Pull the endpoint of every path-op child of <a:path> (moveTo/lnTo/cubicBezTo/ * quadBezTo) into a flat [x, y] list. The endpoint is always the LAST

* 将扁平路径列表转成树形结构 * @param {string[]} paths - 如 ['ppt/slides/slide1.xml', 'ppt/slides/slide2.xml'] * @returns {Object} 树形对象

Instantly pause — tick becomes a no-op.

Per-dimension averages table for a result subset.

* Pick the best visible color from gradient stops. * Skips fully transparent and white colors; falls back to last stop or black.

* Play text using the Web Speech API (browser-native TTS). * Splits text into sentence-level chunks to avoid Chrome's ~15s cutoff. * Uses cancel

* Post-process OMML for PPTX compatibility: * 1. Strip xmlns:w (wordprocessingml is DOCX-only, not valid in PPTX) * 2. Strip redundant xmlns:m (alre

* Post-processing after the agentic loop: * 1. Activate the first issue * 2. Generate initial questions for it using the Question Agent * 3. Add we

* 递归打印树

* Process LaTeX elements: render latex string to HTML using KaTeX. * Fills in html and fixedRatio fields. * Elements that fail conversion are remove

Read a markdown prompt by file name (without extension), cached.

* Read xfrm off/ext from a shape-like node (dsp:sp uses dsp:spPr > a:xfrm).