Research & Data

Organises a group of agents to run a research survey using statecraft from ancient Chinese classics (The Art of War, Han Feizi, Zuo Zhuan, Strategies of the Warring States, Records of the Grand Historian). It uses independent scouting from several entry points, separate verification, and a report graded by evidence level. Use this skill whenever the user asks to survey or 调研 a research area, find research ideas, map related work for a paper, shared task or challenge, scout a field they are new to, or wants to surface things they don't know exist ("unknown unknowns") while keeping out hallucinated or irrelevant results. Use it even when the user doesn't mention agents or the classics, and even for a one-line request like "survey direction X for me" or "帮我调研一下 X 方向".

Organize, format, and index research as OKF-conformant knowledge bundles — dated, sourced, append-only snapshots of the external evidence behind a decision (technology evaluations, library comparisons, industry-practice surveys). Use when recording research output, structuring a docs/research/ session, enforcing source discipline (primary sources, vendor-COI flags, inference-vs-fact labels, confidence levels), or wiring the research → decision → issue traceable chain. Pairs with deep-research (which gathers/verifies the evidence) and living-docs (which owns the ADR/issue artifacts research feeds into).

Write a survey whose every question feeds a decision, with the sample size and analysis worked out first.

Plan and run project-specific ML, VASP and HPC materials calculations from a research goal, including selection, recovery, monitoring and usable results. Use for computational-materials project planning or active calculations; AiiDA is optional.

Draft a short interview guide that gets real stories and tests what you actually need to know.

Map what happens today, or what should happen in future, step by step, from your beliefs or from real research.

Plan a diary study people will actually finish, for behaviour that plays out across days or weeks.

Use when a SQL query is slow, a plan shows Seq Scan or bad row estimates, EXPLAIN ANALYZE output needs reading, an index choice is unclear, OFFSET pagination degrades, an ORM issues N+1 queries, pg_stat_statements needs triage, or a BigQuery/Snowflake query scans too much. Postgres first.

Sort real research notes into the themes people actually share. Needs data to work from.

Guidance for choosing, pricing, licensing, and comparing AI audio models across TTS, ASR, voice conversion, music generation, sound effects, singing synthesis, audio LLMs, and audio restoration. Use when a user asks which audio model or API to use, what it costs at their volume, whether a license allows commercial use, or how to assemble a speech/audio pipeline.

Use when designing, reviewing or debugging batch ETL/ELT in Python - duplicate or missing rows after a rerun, double-counted backfills, watermark gaps, late-arriving data, retry storms, schema drift breaking a load, scrape-to-warehouse pipelines, a job that silently loads zero rows, or choosing between cron, n8n, Airflow and dlt.

Draft, revise, or critique academic prose while preserving meaning, evidence, and the author's voice. Use for research papers, theses, reports, literature reviews, abstracts, and academic paragraphs, including requests to make writing less formulaic, generic, inflated, or AI-like.

Complete guide for data science, statistical computing, graphics devices, ESS integration, and reproducible research using R with Org Babel.

Use for almost any request for facts, research, analysis, comparisons, recommendations, advice, summaries or drafts, including short or casual-sounding ones like \"tell me about X\", \"why did X happen\", \"what's the best X\" or \"find stats on X\". Checks the request for gaps or wrong premises and asks before answering, then holds the answer to cited, verified, complete standards.

将任意研究资料（PDF、Word、Excel、CSV、文本、网页、访谈稿、图片、数据库导出等）转化为可审计的数据集、统计模型、SPSS 风格图表与论文成稿：先做资料完整性、去重与字段质量审计，再按 codebook 编码或抽取，生成 CSV/SAV、跑描述统计与回归、制作 600 dpi 出版级图表，最后组装 Word/PDF 并做数字一致性质检。只要用户提到资料整理、文本编码、问卷/访谈分析、数据清洗、SPSS/Logistic/OLS、实证论文图表、投稿排版，或要在既有数据上追加新材料，就使用本 skill。

Automate best practice documentation workflow - research knowledge base, docs, code, community, synthesize findings, generate document, update INDEX

Institutional-grade Indian stock market (NSE/BSE) intraday, swing, and F&O analysis and deterministic execution engine. Use whenever the user asks to analyze an Indian stock or index (NIFTY, BANKNIFTY, FINNIFTY, SENSEX, or NSE/BSE tickers), evaluate intraday/swing setups (ORB, VWAP, RSI Divergence, Fibonacci), calculate exact INR position sizes with STT/GST/SEBI/stamp duty taxes, run Monte Carlo historical intraday simulations, or stage paper/live orders across major Indian brokers (Zerodha Kite, Angel One SmartAPI, Upstox, Dhan, Fyers, Groww, Kotak Neo, ICICI Breeze, Shoonya).

Audit whether search engines and AI assistants (Google, Bing, ChatGPT, Claude, Gemini) can find, read and recommend a website, then fix what blocks them and grow its reach. Use when the user asks about SEO, AI search visibility (AEO, GEO), whether AI crawlers can access their site, why their site does not show up in ChatGPT or Google, robots.txt or Cloudflare bot blocking, indexing problems, sitemaps, structured data, llms.txt, "is my site agent-ready", the monthly AI answers round, editing their Shopify, WordPress, Webflow, Wix or Squarespace site for SEO, writing pages that assistants cite, outreach to get mentioned, or keyword research.

Use when drafting or revising research writing — papers, lab notes, design documents, research blog posts, related-work sections, or any passage stating findings. Runs a claim audit that makes every assertion falsifiable-as-written, quarantines hedging into a Limits section, and enforces a words-before-symbols ladder for mathematics. Also triggers on requests to review, tighten, or "make this sound like a real paper", and on complaints that a draft is vague, hedged, hard to act on, or reads like AI output.

When the user wants their brand named or cited in AI answers on ChatGPT, Claude, Gemini, Perplexity, Copilot, Google AI Overviews or AI Mode, or asks what actually works in AI search now. Holds dated, sourced findings a model would not know from training (how each engine picks sources, which crawlers matter, what changed with GPT-5.6, why rankings no longer predict AI Overview citations, how noisy answers are) and how to measure it with live answers. Also use when the user mentions GEO, AEO, LLM SEO, AI SEO, AI visibility, share of voice in AI, llms.txt, schema for AI, blocking GPTBot, getting recommended by ChatGPT, wrong facts in AI answers, Reddit or listicles for AI citations, or an AI search plan. Classic Google rankings go to seo, a technical crawl to technical-seo.

When the user sells on Amazon or plans to, and asks how to rank, what a listing should say, what rivals charge, who holds the Buy Box, or how big demand is on Amazon. Holds dated, sourced findings a model lacks (the 75-character title cap and Item Highlights from July 2026, Rufus renamed Alexa for Shopping and picking products that do not rank, the Buy Box eligibility gate removed, reviews split across variations, 2026 fees, Amazon blocking ChatGPT) and how to read Amazon search volume and results data without overclaiming. Also use when the user mentions Amazon SEO, A9 or A10, Amazon keyword research, backend search terms, a listing audit, product titles, bullets or main image rules, an ASIN teardown, rival listings, resellers undercutting us, Vine, getting reviews, Sponsored Products, ACoS, Brand Analytics, an Amazon launch, or a 30-60-90 day Amazon plan. Google Shopping and feeds go to google-shopping, other ad platforms to paid-ads, review mining to voice-of-customer, market sizing to market-research.

Brutally honest co-founder that tries to kill a business or side-project idea with real, sourced research before the user wastes time or money. Extracts a structured idea brief, asks a few sharp questions (country, budget, goal), computes key metrics (market size, CAC, LTV, margin, break-even), scores it out of 100 with a GO / PIVOT / NO-GO verdict, and gives a shareable verdict card and a 7-day test. Use when the user says "kill my idea", "rate my idea", "new idea:", "is this worth building", "should I build X", asks for a go/no-go, or pastes a "make money with X" tweet, post or link.

Help a US homeowner fight a county property assessment: fact packet, public-records request, appeal-board ladder, oversight complaints, refund research, lean media tip. Use when someone pastes an assessment notice or asks to contest an assessor valuation. Not legal advice. Public guide: https://webofmike.com/wa-property-tax-appeal/

Fine-tune an instruct LLM with a LoRA on a local NVIDIA GPU using Unsloth, from a chat-format JSONL dataset, and export it as a GGUF together with a run.json recording what a server needs to prompt it like training did. Use for training or retraining a model, or trying several base models.