Research & Data

Build a chat agent that knows one industry in one metro better than any general model because it carries a precomputed digest of real public records (or a mined review corpus) and can pull live records mid-conversation — finding the moat, building a byte-stable substrate, enumerating what the records can't, the data traps that produce plausible wrong numbers, deterministic math in tools, measured prompt cost and cache TTL, shipping as an ad destination, an off-sector ending, and a precomputed spatial "what is this next to" layer. Use when building a vertical AI agent grounded in records rather than general knowledge.

Acoustic fly-over geolocation: locate where an audio or video recording was made from the near-inaudible Doppler curves of propeller aircraft and helicopters in the background, matched against historical ADS-B tracks (about 100 m in good conditions), with an honest uncertainty region. Also covers thunder plus lightning-data localisation and a pre-publication acoustic privacy self-audit. Use this skill whenever the user wants to locate, verify or time a video, recording, livestream or call from its sound, background noise, aircraft noise, Doppler shift, spectrogram, ADS-B data or thunder; mentions acoustic geolocation, audio OSINT, "where was this recorded", verifying a video's location by sound, or colsto's method; or wants to check whether their own video leaks its location through sound, even if they never say Doppler or ADS-B. For OSINT verification, journalism, CTFs, research and self-audits; ships a tested Python pipeline.

Family history and genealogy research done like a trained historian. Use when the user works on ancestry, a family tree, a family archive, old photos or letters of relatives, or searches for a person in archives, church books (Kirchenbücher), civil registers (Standesamt, état civil), newspapers, address books, military or prisoner records, emigration lists, or Nazi-era records of ancestors (NSDAP card file, Wehrmacht, persecution). Works in German and English, with deep coverage of German, Central and Eastern European sources and solid international coverage. Enforces source criticism, citations for every claim, explicit confidence levels, documented negative results, privacy of living people and careful handling of the Nazi era.

Get a specific record out of a US government office by public-records request — "request the assessor's model", "file a data practices request", "FOIA this", "open records request", "can we just ask the county for it", "who do I write to at <county>", "chase that request", "did they ever answer". Covers checking the website first, finding the right office and statute, the carve-out paragraph that makes a request answerable instead of refused, web-form traps (hidden character caps, WAFs that reject apostrophes), chasing (there is usually no deadline), whether what comes back may be republished, and parsing multi-year tables without silent errors.

Use for any \"research and plan X\" request. Triages source access first, plans only from what was actually read, and writes plan.md plus a categorized, topic-named sources file. Re-run after downloads to update.

Audits any research output (plan, report, lit review, draft) for fabricated references, claims a source doesn't support, abstract-only claims and unbounded \"no one has studied X\" claims. Run it before delivering research.

Finds sources on a topic and classifies each one before anything is read: legal access, which route an AI agent can actually read, integrity, and grade. Writes a ledger the other research skills use.

Assembles a complete LinkedIn outbound campaign brief into campaign-NAME.md - segment, signal stack, list-building steps in Apollo and Sales Navigator, the opener and follow-up sequence, LinkedIn tool setup notes (SalesRobot or similar) with safe daily limits and warm-up, metrics with targets, and a 14-day plan. Use when the user says "build a campaign", "plan my outbound", "set up a LinkedIn campaign", "campaign brief" or "launch a campaign for a segment".

Evidence-driven application security audit and hardening for repositories you own or are authorized to assess. Detects the stack, inventories the attack surface, runs SAST/SCA/secrets/IaC/supply-chain checks (built-in rules plus Semgrep, Gitleaks, Trivy, OSV-Scanner and others when installed), performs threat modeling and manual secure-code review against OWASP Top 10, API Top 10 and ASVS, validates findings safely, and writes Markdown, JSON and SARIF reports with a remediation roadmap. Use when

Guide a student from zero to a finished scientific article or thesis with responsible AI help, following the Persian course «از صفر تا صد نگارش و انتشار مقاله علمی با کمک مسئولانه هوش مصنوعی» - a ten-folder dossier, 18 parts with pass criteria (ایستگاه), right-to-left Word forms, the course's session handouts and prompt book, path C for a study with the student's own questionnaire data (design, sample size, ethics, Method, cleaning, analysis, Results, Discussion) and a thesis track (proposal, ch

Considered, well-researched advice on a big or nuanced marketing-leadership question. Gathers company and personal context, searches the whole vault, runs deep web research, and tailors the answer to the owner's situation. Invoke with /advise-deep <question> [workspace:<company>], or /advise-deep setup to change its settings.

Count banned design patterns in UI work and report the tally before presenting anything. Use for any anti-slop, de-slop, or "avoid AI slop" request; for design lint, audit, critique, review, or cleanup; and whenever producing or revising UI — screens, pages, components, mockups, design code, CSS/Tailwind/JSX, or Paper/Figma artboards — including redesigns, polish passes, and multi-variant explorations. Triggers on "avoid ai slop", "no slop", "deslop", "is this generic", "does this look AI-made",

Research a Reddit audience with KissMyPost data. Use when the user asks which subreddits fit a product or topic, how big or active a subreddit is, when to post there, which communities share its audience, what Reddit talks about around a topic, or how often a brand or phrase is mentioned on Reddit.

Interview a founder, gather product and GTM evidence, evaluate all 19 Bullseye marketing channels, and design three focused two-week acquisition tests. Use when choosing or reassessing acquisition channels with the Bullseye framework.

Revise user-provided English scientific-writing paragraphs using a structured workflow: identify the manuscript section and paragraph function, improve scientific clarity and evidence calibration, explain each major change, and provide a polished revision. Use for academic English revision, not general translation or invented scientific content.

Research family history, interpret existing trees and documents, distinguish namesakes, and build source-cited kinship trees, especially for the former USSR and Russian Empire. Use for ancestry, historical archive searches, conflicting relationships, and GEDCOM reports; not current-person tracking.

Use ao construir, medir ou revisar corpus bibliográfico, busca em bases (PubMed, OpenAlex, Crossref, DOAJ, ERIC, SciELO), contagem, interseção, lacuna, triagem, cocitação, coautoria, thesaurus, Zotero. Contém as armadilhas que produzem conclusão errada: número certo pelo motivo errado. Consulte ANTES de afirmar que algo não existe na literatura. Para desenhar o mapa depois de ter os dados, veja a skill vosviewer.

Validate an app idea against the App Store before building it. Turns the idea into search phrases, measures demand (App Store autocomplete) and competition (current top-10 apps) for each, ranks keywords by opportunity, and maps which apps own the market. Use when the user wants keyword research, ASO, "is there demand for…", "who are the competitors for…", "what should we call it / which keywords", or to validate or compare app ideas.

Scans current signals across UI/UX, product design, design strategy, AI-assisted workflows, the design toolchain, B2B/B2G/SaaS design, AI regulation and process automation, then generates content angles calibrated to the user's own positioning. Use when the user asks for content ideas, wants to know what is trending or debated in design, product or AI, needs an editorial plan, or names a topic to write about (search first, then differentiate).

分析 Apache Spark / Spark SQL 任务执行情况。根据 SQL 原文、EXPLAIN FORMATTED、Kyuubi/Spark Driver 运行日志和 Spark UI 地址，定位慢查询瓶颈、数据倾斜、spill、GC、调度/冷启动、join 膨胀和配置问题。触发词：Spark UI、SparkUI、Spark SQL 分析、SQL 慢、数据倾斜、stage 慢、task 长尾、Spark 任务诊断。

Converts CSV/Excel datasets and R, Python, or SPSS statistical outputs into publication-quality statistical visualizations and multi-panel analytical dashboards, emitting standalone reproducible Python figure scripts and SHA-256 data provenance manifests.

Answer questions that need Belgian public data (population, municipalities, companies and the KBO/BCE, employment, mobility, energy, environment, elections, addresses, parcels) by finding the official source, checking its licence and freshness, and reading the real file structure before using it. Use when a user asks for Belgian statistics or datasets, or wants to join, compare or cite Belgian data.

Builds Power BI semantic models and reports from a dashboard spec — star schema in TMDL (PBIP project format), DAX measures with time intelligence and semi-additive patterns, report pages in PBIR, themes, drill-through, field parameters and row-level security. Use this skill whenever the user wants a Power BI dashboard, report, semantic model, dataset, DAX measures or a .pbip project built or changed, including when a domain skill (finance, supply chain, healthcare, sales, HR) has produced a dashboard spec and the target platform is Power BI or Fabric.

Use when deciding how a software team should work, especially once AI agents write much of the code. Covers how changes get reviewed, approved, merged and rolled out; how to test or verify work, including evals, golden datasets and LLM graders; how to write tasks, CLAUDE.md or REVIEW.md for agents and pick effort or session type; how to run agents unattended or on a schedule; how to design an agent's tools and context; and how to measure engineering cost and quality. Also use when asked how Anthropic or the Claude Code team does any of these.