tashanwork

Daytona cloud instance, Den server, OpenWork Cloud, Marketplace onboarding. Use when the user asks to run, launch, start, validate, or record a Daytona cloud/Den instance for OpenWork Cloud flows.

Daytona cloud server and Den sandbox setup. Use when the user says Daytona server, cloud server, Den server, marketplace server, worker proxy, cloud auth, org policies, or connect Electron to a Daytona server.

Validate Electron against a Daytona Den server. Use for two-sandbox cloud auth, marketplace, org policy, worker proxy, provider sync, or desktop handoff flows.

Daytona Electron sandbox testing with CDP/noVNC. Use when the user says test on Daytona, run Electron on Daytona, Daytona dry run, test Electron remotely, reproduce on Daytona, or validate a real desktop flow.

Daytona UI flow validation loop. Use when validating real app behavior, checking a Daytona flow, proving a bug is fixed, or deciding pass/fail from CDP snapshots, screenshots, and assertions.

Daytona recording volume, screenshots, artifacts, and validation evidence. Use when the user says record Daytona, recording volume, artifacts volume, screenshots, proof, PR evidence, before/after video, or validate behavior visually.

Daytona secrets volume setup. Use when the user says Daytona secrets, add provider key, OpenAI key, Anthropic key, eval secrets, /daytona-secrets, or openwork-eval-secrets.

Run OpenWork UI evals on a Daytona sandbox or local Electron instance. Handles sandbox creation, service startup, and eval execution via CDP browser tools.

用于判断文本的 AI 痕迹、模板腔、过度均质化表达和不自然写作风险，并给出保真改写建议。适用于“AIGC监测”“AI味检测”“不像人写”“去模板腔”“客户/论文/方案文本润色但别改事实”的场景。

在对话里慢慢形成一份可修正的用户画像。用于用户要求建立、查看、更新画像， 希望回答更贴合自己，或反复纠正"不像我""太官方""理解错了"的场景。 也用于用户说"记住""以后默认""按我之前习惯""别再这样""不要记录" "删掉""恢复画像""你记了什么""你忘了"，或普通任务里需要沿用已知偏好的场景。 只记录会影响后续回答的偏好、判断习惯、表达边界和待确认的地方。

Call and verify the Deep Research SSE API for research questions, literature-backed reports, keyword planning, private paper search, references extraction, and evidence-boundary reporting. Use when Codex needs to run or dry-run Deep Research, inspect SSE phases, diagnose empty references or incomplete answers, or produce a source-aware research report from the Deep Research endpoint.

调用或构造 Giiisp/集思谱论文检索请求，并按任务路由到 arXiv、OpenAlex、CNKI、跨平台去重、引用网络、期刊检查和引用审计。用于文献检索、文献初筛、arXiv 定位、中文论文检索、开放源补充和引用核验。

使用 Giiisp Imagine 科研图像生成接口，把论文段落、图题、实验流程或机制描述整理成作图简报、生成请求、图片结果和检查记录。适用于论文配图、科研流程图、机制通路图、实验示意图和学术汇报图。

用于检查回答、报告、方案或资料摘要中的幻觉风险、无证据断言和过强结论。适用于用户要求“查幻觉”“核验事实”“看看哪里不可信”“避免编造引用/数字/结论”的场景。

Use when Codex needs to create, run, review, debug, or package Manim Agent workflows for mathematical or technical explainer videos: Manim animations, narrated teaching videos, formula or proof visualization, concept explainers, render review, DashScope CosyVoice TTS, FFmpeg muxing, or the local gqy20/manim-agent CLI/Web pipeline. Trigger on requests such as "做一个 Manim 动画", "生成讲解视频", "把这个公式可视化", "做数学动画", "带配音", "渲染成 mp4", "检查 Manim Agent", or "包装/维护这个 manim-agent skill".

Run no-key citation audits for academic papers using local evidence extraction, bundled TaShan-PaperChecker rules, and the current Codex model. Use when Codex needs to check .docx论文引用合规、正文引文和参考文献是否匹配、引用真实性、引文与上下文相关性、GB/T 7714 or UCAS reference-format issues, or when the user mentions PaperCheck, TaShan-PaperChecker, citation compliance, reference audit, citation authenticity, or 引文核验.

Translate scientific questions into clear data inputs, outputs, and baseline workflows. Use when Codex needs to help researchers clarify what data goes in, what target or result should come out, how samples/labels/features are defined, and how to solve the resulting data problem through visualization, preprocessing, baseline design, training, evaluation, and reporting.

用于中文科研文本的 scientific humanization：把论文、综述、基金、项目书、技术方案、答辩、课程讲稿、PPT讲稿和投资人科研表达改得更像专业作者或讲师本人在说，同时保护事实、术语、数字、引用、图表编号和证据强度。适用于用户说“别AI味”“人一点”“别太官方”“像我讲的”“去模板腔”“改顺但别改事实”的场景。

Generate evidence-tracked research ideas through an arXiv-based, skill-native Scispark workflow. Use when Codex needs to turn a research keyword, question, paper set, Zotero/library material, or arXiv results into structured facts, testable hypotheses, an initial research idea, technical optimization, mechanism-of-action analysis, human-AI collaboration review, or optional academic slides. Also use when the user mentions Scispark, 科研想法生成, 研究假设生成, 机制优化, MoA, literature-backed idea generation, or 从关键词到研究方案.

一键接入他山世界，获得专属科研助手；支持数字分身获取资讯并根据反馈迭代，参与虾们的头脑风暴与科研社交，进入 Arcade 公众科学竞技场，并在 SkillsHub 中自发挖掘、安装和评测科研技能。

Build image-model-driven PPT decks from a user topic, source materials, papers, reports, notes, datasets, existing slides, or style references. Use when the user asks to create, redesign, package, or QA a PowerPoint deck where every slide should first be generated as a high-quality full-slide image and packaged into PPTX with manifests, previews, and evidence-backed QA. Layered editable reconstruction is optional and only used when explicitly requested.