Fan out gen-image over every pending slot in asset_plan.json.

Fan out the standalone `refine-page` workflow over every page that has a built HTML file. Each per-page task does its own screenshot → VLM cri

Generate a single slot's image via sn-image-base's sn_agent_runner (T2I). Policy: T2I must route through sn-image-base, NOT through model_client.

列出 PMC 论文所有章节目录。

列出论文所有章节（不含正文）。

列出 DeepXiv 已解析的论文元数据和章节结构。

Two-step HTML generation: Step 1 (REWRITE): convert the structured outline + style + inherited content into a single natural-language "quer

读取整篇论文的正文内容。

读取 DeepXiv 返回的整篇 markdown 正文。

读取指定章节的正文内容（含子章节文本）。

读取指定章节的正文内容。

读取指定章节正文。

Three-step page refinement (screenshot → VLM critique → LLM apply). Outputs (per page): - `screenshots/page_NNN.png` (rendered slide image

Returns (pool, alias_map, claim_to_sources): pool[primary_id] = Citation alias_map[any_id] = primary_id claim_to_sources[claim_

Create an httpx.AsyncClient with shared defaults for vigeneval evaluators. Automatically uses proxy from environment variables (HTTPS_PROXY, HTTP

* Detect whether a bar-type chart is horizontal. * ECharts convention: horizontal bars have yAxis.type='category' + xAxis.type='value'.

* Pick the "primary" type. If multiple series with mixed types, return 'combo'. * Special cases: * - line + areaStyle → 'area' * - pie wit

Detect image type from filename and optional context text.

Infer MIME type from image magic bytes. Args: data: Raw image bytes (at least 8 bytes for PNG check). Returns: str: 'image/p

Download an image from a URL. Args: url: The URL of the image to download. timeout: The timeout for the request. Returns:

下载论文 PDF。隐藏兼容接口，不在 skill 使用说明中公开。

* 首次运行时自动安装依赖（npm install + playwright chromium）。 * 后续运行检测到 node_modules 和 chromium 已存在则跳过。

* 提取元素的边界框（相对于 .wrapper 左上角）。 * @param {Element} el * @param {DOMRect} wrapperRect * @returns {{x:number, y:number, w:number, h:number}}

* Doughnut/pie center label — pulled from option.graphic[].style.text or * series[0].label (when position='center'). The caller renders this as a *

* Extract series-level data label color (from series[0].label.color).

* Extract legend text color.

* 提取列表数据。 * @param {HTMLUListElement|HTMLOListElement} listEl * @returns {Array<{text:string, styles:Object}>}

* 提取元素的 ::before / ::after 伪元素。 * 仅当伪元素有可见内容（背景、尺寸）时提取为合成 IR 节点。 * @param {Element} el * @param {DOMRect} wrapperRect * @returns {Array<Ob

* 提取表格数据。 * @param {HTMLTableElement} table * @returns {Array<Array<{text:string, isHeader:boolean, colspan:number, rowspan:number, styles:Objec

* 提取混合内容的 textRuns 数组。 * @param {Element} el * @returns {Array<{text:string, bold:boolean, italic:boolean, fontSize:string, color:string, fontFa

Recover a JSON value from raw text, or raise ValueError if none found.

Return extracted text and page count, preferring pypdf and falling back to pdftotext.

Extract the number shown after the quote icon/marker in a paper card.

通过 ID 列表直接获取论文元数据（不做文本搜索）。

获取论文的 references 或 citations。

获取论文的 references 或 citations：SDK 优先，失败时使用原 HTTP Graph API。