Research & Data

Build a grounded literature corpus for a scientific hypothesis and publish it into a Symposium community record. Takes one hypothesis stated "If [Cause], then [Effect], because [Rationale]", searches the literature and data repositories for the sources that bear on it, expands one citation hop, resolves access, puts the corpus to the user for review, then hands it to Symposium to publish as ScientificPublication, Data and Argument artifacts. Use for: curate sources for a hypothesis, build a literature corpus, find papers supporting or contradicting a hypothesis, import scientific publications into a community record, gather evidence for a Symposium record.

Turn a plain-English stock strategy into a validated strategy definition, backtest it only after the user confirms, explain the results using only numbers from the tool output, and start paper trading on explicit approval. Runs only the openclaw-bot commands.

Synthesises evidence from multiple sources into a decision-ready, fully traceable report. Frames the question, weighs each source, records evidence in a claim-level ledger, builds claims across sources (thematic, narrative, framework, quantitative or mixed), computes how certain each claim is allowed to be, stress-tests the conclusions and verifies every citation. Use when asked for a research synthesis, evidence synthesis, literature review, systematic, scoping or rapid review, evidence summary or briefing, "what does the research say", the state of the evidence, or to combine, compare or reconcile papers, studies, reports, interview transcripts, survey responses or other research data. Also use to audit or strengthen an existing review. Not for summarising a single document or for quick factual lookups.

Use when a user asks what the EU Cyber Resilience Act (CRA) means for their product, whether and when they must report a vulnerability or incident, or what a specific CVE triggers legally. Maps a product with digital elements to CRA scope, product classification, Annex I vulnerability-handling duties, and Article 14 reporting obligations — every legal claim cited from official regulation text fetched live through the Ansvar Gateway MCP connector, joined with live CVE / CISA-KEV / EPSS vulnerability intelligence from the same connector.

Evaluate a research, product or start-up proposal, experiment report or paper draft against Paul Graham's "How to Do Great Work" standard. Scores five dimensions on a -1 / 0 / +1 scale (true and new, honesty and rigor, generality, quality of the question, first step and value of failure), searches for the closest existing work, checks the strongest simple baseline, and fact-checks the author's own answers on scope, curiosity and "why you" when they are submitted alongside. Use this skill whenever someone asks to review, score, critique or get feedback on a proposal, project idea, paper plan, grant draft or early report, asks whether an idea is novel or worth pursuing, or mentions "great work", even if they do not name the skill.

Turn natural-language circuit requirements into a complete, evidence-led electronics design. Elicit critical specifications and preferences, make only conventional low-risk assumptions, research and select real components, calculate and simulate the design, and deliver a KiCad project with schematic and PCB, assigned footprints, BOM, SPICE netlist, and verification evidence. Use for new circuit design, circuit redesign, component sizing, KiCad/SPICE generation, simulation, and design verification.

Bioinformatics literature systematic review skill. Use when interpreting a scientific paper PDF in bioinformatics, genomics, oncology, or related fields. Extracts full text AND every main figure as high-resolution PNGs (PyMuPDF caption-anchored region rendering), then generates standardized deliverables: a Word deep-reading report (10-section structure with each figure embedded and analyzed: research question / method / key results / interpretation), a Nature/Cell-style PPT for group meeting sharing, and optionally a popular science article for WeChat Official Account. Covers the full workflow from PDF parsing, figure extraction, deep reading, structured analysis, to document generation.

Interview the user about a scientific paper they have read, asking probing questions like a fair examiner so they can check and deepen their own understanding. Use when someone wants to be quizzed, questioned or examined on a paper, prepare for a journal club or exam, or test whether they understood a paper or one of its sections.

Sizes a startup's market – builds TAM/SAM/SOM twice (bottom-up from buyers × price, and top-down from a published market figure), triangulates the two, then builds a third number the demand builds cannot give you: the capacity ceiling their own sales and delivery machine can carry. Compares all of it to whatever the deck claims, and reports the distance between what the market could support and what the company's own machine can reach. Produces a sizing note, one csv of every line item, and one chart. Bootstraps the company folder and researches the source material itself when none exists. Invoke when the user says "size the market for X", "what's the TAM for X", or "/market-size X".

Turns a startup's revenue data into MRR (monthly recurring revenue), cohorts, NRR (net revenue retention) and one investor slide. Reads monthly update files, a per-customer revenue export, or a contracts register (client, contract value, start month, term). All arithmetic is done by a tested Python helper, never in prose. Separates billed months from contracted-but-not-yet-billed months, reports contracts shorter than three months as non-recurring, and says "cannot compute yet" for a 12-month NRR when no customer has run twelve months. Produces a plain-English note, four CSVs and one 16:9 slide. Invoke when the user says "build the metrics pack for X", "MRR, cohorts and NRR for X", "what is X's net revenue retention", or "/metrics-pack X".

Analyze image classification datasets before training for visual quality issues, semantic outliers, class confusion, augmentation instability, corrupt or duplicate samples, relative difficulty bands, and stratified splits. Use for dataset audits and training-readiness checks; do not use for detection, segmentation, video, or claims about downstream accuracy.

Designs leakage-safe scikit-learn splits, cross-validation, scoring, search, diagnostic curves, PCA scaling, model inspection, ColumnTransformer, and Pipelines. Use when building or reviewing an sklearn evaluation workflow. Do not use for ordinary prediction with an already-fitted estimator, plotting already-computed predictions, experiment tracking after evaluation design is fixed, deep-learning frameworks, estimator theory alone, or incidental sklearn imports.

Maps a startup's competitors – writes an explicit membership test, sweeps for operators through several independent routes, classifies each as direct / former direct / indirect, compares pricing like-for-like from published pages, researches total funding per competitor with an entity check, and sweeps M&A comparables with deal values and multiples. Produces a competitor note, one csv, one price chart, and a two-sheet shareable market-analysis workbook (competitors + M&A comparables). Bootstraps the company folder and researches the source material itself when none exists. Invoke when the user says "who competes with X", "competitor research on X", or "/competitors X".

使用 Kaiwu PyTorch Plugin（KPP）安装、构建与训练 PyTorch 能量模型，并接入 Kaiwu 采样器。适用于 RBM/BM、DBN、使用能量模型先验的 QVAE、能量引导离散序列生成 QDiffusion、为 PyTorch 模型做 MAIFS 特征选择，或 kaiwu.torch_plugin 的调试。明确需要 KPP 或场景适配时主动检查并安装 kaiwu-torch-plugin、处理依赖冲突、验证导入和训练。普通回归、一般神经网络、普通 VAE/扩散模型不触发；不依附 PyTorch 模型的纯 QUBO/Ising 组合优化不属于本 Skill 的范围。

Discover and compare open-source projects and tools for a repository, local project, idea, or learning goal, including graduation projects and AI creative workflows. Rank learning and reuse value with source evidence and produce a standalone HTML shortlist.

Lee un texto filosófico o teórico, extrae su estructura conceptual y la dibuja como diagrama TikZ de calidad editorial, en blanco y negro, fiel al texto y distinguiendo qué tipo de relación une cada par de ideas (derivación, génesis, reciprocidad, unilateralidad, retorno, totalización). Úsala siempre que se pida «el diagrama de este texto», «grafica las relaciones entre los conceptos de…», un mapa conceptual o esquema de una obra o capítulo (Kant, Deleuze, Laruelle, Bergson, Aristóteles, Platón, Carnap, Sneed…), «pásalo a TikZ», «visualiza el argumento», o cuando se comparta un PDF, imagen o fragmento filosófico con un pedido de figura, aunque no diga «diagrama» ni «TikZ». Reconstruye la arquitectura con un libro de fidelidad (cada nodo y arista con su pasaje) y un léxico donde la forma del trazo lleva el significado. No es para gráficos de datos, animaciones o videos (html-concept-animations, manim-concept-videos, claude-motion) ni para diagramas de software o de flujo.

Draw publication-quality hardware/FPGA/dataflow block diagrams in TikZ (HLS kernels, AIE arrays, systolic arrays, SoCs, pipelines) in greyscale or colour-blind-safe palettes.

Identify installed or removed fire sprinklers from phone photos, readable markings, sprinkler identification numbers (SINs) or records, check U.S. recalls and replacement warnings, and screen markings against UL and FM Approvals counterfeit and falsified-certificate alerts. Use when a user uploads a sprinkler photo, asks whether a sprinkler is recalled or counterfeit, asks if a UL or FM mark is genuine, requests identification of Central, Omega, Gem, Star, Globe, Tyco, Viking or other sprinklers, asks to distinguish O-ring and Belleville seals, asks about O-ring corrosion risk, or shares a sprinkler certificate. Return a short evidence-based screening result, flag O-ring corrosion risk and counterfeit-alert cues, and request specific missing views without guessing. Informational screening aid only, not an inspection, engineering opinion, product authentication or code compliance determination; see DISCLAIMER.md.

Use when a course goal, its weekly goals and its success criteria (an h0 inputs.json) must be broken into a capability tree of what participants must DO with Claude (h1 tree.json). Builds evidence-up: starts from what must be observably true and works back to the goal. Writes three gate-passing versions (A/B/C), pitches, and a rendered leaf list.

输入任意公司名称，先做主体定位与歧义消解，再输出结构化公司画像（主营/次要业务线、近5个财年营业额、产品与品牌、所属国家、中国区员工数与投资情况），每项标注来源与数据截止时间，缺失项显式标注不臆测。当用户说「介绍一下XX公司」「XX公司是做什么的」「XX营收多少」「查一下XX这家公司」「XX在中国的员工和投资」「XX属于哪个国家」或只发来一个公司名时触发。

Teaches so the user can understand it or do it themselves afterward, instead of handing over answers. Use whenever someone wants to learn: teach me, explain, ELI5, walk me through, help me understand, quiz me, flashcards, "I'm rusty on", "what is X", or a bare concept name ("Galois theory", "transformers from scratch"). Use for learning a skill (git rebase, SQL joins, guitar chords, a codebase "so I can work in it"), and for any request to build a lesson, course, study guide, practice set or interactive artifact to learn from. Also use on confusion signals ("won't stick", "keep mixing these up", "not getting it") and learning-path questions (prerequisites, what to study first). Don't use when the user wants the result rather than the skill (write or fix this code, draft this email, translate, calculate, look up a fact, news), for personal troubleshooting, for book or course recommendations, or for Claude's own verdict ("do you think X", "honest take", "settle this").

Build a BlenderGIS terrain map and grayscale elevation render from user-provided coordinates, Map Range bounds, and output name. Use for this georeferenced basemap-to-SRTM workflow, not unrelated Blender work.

将按顺序排列的教学视频及其字幕、文稿和画面整理成可操作的详细跟练材料。适用于用户希望逐步复现网课、核对过程效果和理解老师补充知识的场景；不用于仅凭课程标题或简介概述课程。

Measure a web page, local HTML file, game recording or video for photosensitive epilepsy (seizure) risk and report against WCAG 2.3.1 (Three Flashes or Below Threshold). Use when the user asks to check, audit or test anything that flashes, flickers, strobes, blinks or has fast animation for epilepsy, seizure, photosensitivity, WCAG 2.3.1 or accessibility safety.