Agent Skills

---

name: firecrawl-crawl

description: Bulk-extract many pages from one site or section. Use for "crawl", "everything under /docs", or content spanning linked pages.

---

name: firecrawl-developer-index

description: Search an index of public repositories, GitHub issues, merged pull requests, repository READMEs, and curated documentation sites. Use when a programming question needs external documentation or upstream evidence.

---

name: firecrawl-download

description: Save a site or section as local files (markdown, screenshots). Use for "download the site", offline docs, or a local copy for reference.

---

name: firecrawl-interact

description: Drive a live browser on a scraped page: click, fill forms, log in, paginate, infinite-scroll. Use when content requires interaction or a scrape failed or returned incomplete content.

---

name: firecrawl-map

description: Discover and list a site's URLs, with search filtering. Use for "map the site" or "find the URL for" requests — when the user knows the site but not the exact page, or wants site structure.

---

name: firecrawl-monitor

description: Alert by webhook/email on web changes — use for "monitor/watch/track/alert me when": recurring checks on known URLs (prefer over repeated one-off scrapes) or web-wide watches for new results (queries + goal).

---

name: firecrawl-parse

description: Convert a local file (PDF, DOCX, XLSX, HTML, …) to markdown, or answer questions about its content. Use whenever the input is a file path, not a URL.

---

name: firecrawl-research-index

description: Find the papers that answer a research query in Firecrawl's research paper index — a corpus of paper abstracts whose largest share is biomedical and life-science literature (PubMed, bioRxiv, medRxiv), alongside arXiv preprints in CS, physics, and math — using semantic search, semantic and structural expansion, and in-body verification. Use this skill for literature-finding and paper-retrieval tasks of any kind, including clinical, biomedical, drug, gene, disease, and other life-science questions, whether the answer is a single paper or a full multi-paper set. The index is reached only through the `firecrawl_research_*` MCP tools or the `firecrawl research` CLI subcommands. Calling `firecrawl_search` with its `categories` option set to `["research"]` is a different feature — it filters ordinary web search to research-affiliated websites (the list includes PubMed, bioRxiv, medRxiv, arXiv, and publisher sites) and returns page results from them, without querying the paper records in this index.

---

name: firecrawl-scrape

description: Read a known webpage or execute a discovered workflow or data-provider capability. Use for page content or structured results once the URL or tool is selected.

---

name: firecrawl-search

description: Find web sources with query-relevant page excerpts and optional full-page content, and discover workflows, data APIs, and indexes. Use for web research or finding structured records, listings, transcripts, and datasets. Supports semantic tool discovery, domain matching, and progressive catalogue browsing.

---

name: qu-ai-wei

description: Rewrite and humanize Simplified Chinese while preserving facts, meaning, evidence strength, register, and authorial voice. 用于 `/qu-ai-wei`、「去 AI 味」「改得说人话」「humanize 中文」「改自然点」「改写 / 重写这段中文」「润色得自然些」「太生硬了」等请求，包括段落、文章与长文的结构重写。不要用于翻译、新写中文、只查错别字、繁體中文，或未经授权替真人更换声口。

---

name: agent-browser

description: Browser automation CLI for AI agents. Requires Node.js with agent-browser. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.

---

name: calculator

description: A simple calculator that can add, subtract, multiply, and divide numbers. Use when the user needs to perform basic arithmetic operations.

---

name: database-query

description: 通过 Evotown Database MCP Proxy 对已注册的业务数据库执行只读 SQL 查询。当用户需要查 CRM、订单、报表等结构化数据时使用；禁止直连数据库或写入 SQL。

---

name: evotown-dispatch-complete

description: Report Evotown dispatch job completion after the agent finishes real work.

---

name: evotown-handoff

description: Queue a cross-agent or cross-team handoff through the Evotown control plane.

---

name: find-skills

description: Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.

---

name: http-request

description: 发起 HTTP 网络请求，支持 GET、POST、PUT、DELETE、PATCH 方法。当用户需要调用 API、获取网页内容、发送数据到服务器时使用。

---

name: skill-creator

description: Create or update AgentSkills. Use when designing, structuring, or packaging skills with scripts, references, and assets.

---

name: light-citation

description: Verify scholarly references and claim-citation support for Light stage 10. Use when auditing a manuscript, claim map, bibliography, DOI/arXiv/PMID/ISBN/URL, BibTeX/CSL, citekeys, chimeric or fabricated citations, retraction/correction alerts, or preparing a canonical citation registry for typesetting. Builds provenance-preserving inventories, confirms metadata with independent authoritative sources, distinguishes CONFIRMED/CONFIRMED-MISSING/UNAVAILABLE/UNRESOLVED, records Crossref update direction, and emits the citation gate plus delivery artifacts.

---

name: light-consistency

description: Light 跨材料一致性常驻机读门:在所有产出材料后台核查术语 / 指标(名+值)/ 创新点表述 / 方法名 在论文·PPT·软著·代码·项目文档之间是否统一,定义一改回扫所有已产出材料,把"各处说法对齐"从口头 建议落成**可机检、可阻断、可被总控 run_checkpoint 聚合的机读门**(产 light.findings.v1,术语/指标/ 创新点硬冲突 → Critical fail → exit 1)。何时用:任一材料产出或修改后、投稿/答辩/提交前的跨材料回扫、 受控定义(术语/指标/创新点)变更后。触发词:一致性 / 术语统一 / 指标对不上 / F1 vs 准确率 / 数值冲突 / 论文和PPT不一致 / 创新点表述 / 方法名 / 改名 / 软著与系统 / 术语表 / 回扫。核心纪律:事实源是项目 `.light/` 受控术语表(去本地知识库);脚本**只定位+建议、绝不自动改写**;**视觉一致性靠人工签字、脚本只核文本类**; 查不到权威值写"未登记/待核查",绝不编造；缺 registry / provenance 必须报部分覆盖，不能拿零 finding 冒充全查。

---

name: light-data-engineering

description: Light 科研主线第 2 步·数据工程：**找得到且用得起的数据**（来源/许可/版本/大小/split）+ **提 idea 前先判数据可行性** （数据够不够支撑研究/统计功效）+ **防数据泄漏**（顶会拒稿高频雷）。何时用：用户要找/选/下载公开数据集，或给了数据问 "能不能做研究/够不够/质量行不行" / 要清洗·处理缺失异常·特征工程·划分数据集·数据增强 / 自建数据集（采集·标注规范· 隐私合规·发布） / 怀疑训练测试串了数据(泄漏) / 提 idea 前评数据基础。 触发词：数据够不够 / 数据可行性 / 数据质量 / 数据泄漏 / 防穿越 / train test 重叠 / 怎么划分 / 交叉验证 / 标注规范 / 一致性 IAA / 自建数据集 / 样本量够吗 / 统计功效 / 找数据集 / 数据许可 / dataset search / data leakage / feasibility / data split / annotation。核心纪律： **数据泄漏 = critical 一票否决**（标准化早于划分/时序穿越/实体重叠/目标编码穿越）；**数据不足以支撑 idea = 拦在 idea 前（回边 2⊣3，补数据/改 idea）**；功效是经验阈值非 power analysis；泄漏检测是启发式有边界，不吹"查全了"。

---

name: light-experiment-coding

description: Light 科研主线 stage 6：把冻结的 question/estimand、experiment matrix、pre-registration 与 data lineage 落成最小可运行、测试先行、无泄漏、可复现且能交给 result-analysis 的实验代码。用于实现或复现训练/预处理/评测， 设计 gold/property/metamorphic 测试，控制 Python/NumPy/框架/CUDA/DataLoader 随机性，审查 train/test 或 CV fit 穿越，记录 config/code/environment/input hashes、stdout/stderr、raw metrics、patient/entity predictions 与 failure artifacts，以及运行 stage-6 checkpoint。数据泄漏或不可复现是 critical；静态扫描和同 seed 两次一致都不证明跨硬件 绝对复现。

---

name: light-figure

description: Light 科研主线第 9 步·图表：图服务论点（每图支撑哪条 claim、删了缺什么）+ 出版级规格（栏宽/字号/色盲安全/ 误差棒+n+显著性）+ 视觉诚实（不偷偷截 y 轴/不双 y 轴伪相关/不 jet-rainbow）+ 渲染后多模态「真看一眼」+ 论文数据图程序化生成绝不 AI 生图。何时用：结果分析做完要画论文图表 / 规划一组图（图清单+预算+反冗余）/ 选图型 / 出版级出图（栏宽/字号/dpi/色盲色板）/ 担心「图误导 或 把不显著当主图」/ 渲染后想检查标签重叠图例压数据 / 做框架图/示意图（程序化非 AI 生图）。 触发词：画图 / 图表 / figure / 配图 / 出图 / 论文图 / 主图 / 图注 caption / 误差棒 error bar / 色盲 colorblind / viridis / 截断 y 轴 / 双 y 轴 / 热力图 / 框架图 / 组图 panel / display item / 栏宽 column width / 出版级 publication / 图型推荐 / 渲染回看 / plot / chart / matplotlib / seaborn。 核心纪律：**视觉不诚实(截 y 轴/双 y 轴伪相关) = critical 门**（spec §4.2，STAGE_GATES[9]=[visual_honesty]）； 误差棒缺失 / jet-rainbow / 集合超预算 / 反冗余 = warn；**把不显著结果当主图/图文不一致 = critical → 9→7 回炉发起**； **论文数据图必须程序化生成（matplotlib/seaborn/R），绝不 AI 生图**（永久底线）；静态 lint 只抓形态可疑，真误导终判需图注+人/审稿人。 本技能是 **9→7（把不显著当主图/图文不一致→result-analysis）回炉发起方**。