데이터·AI 스킬

Transform vague inputs into precision-optimized AI prompts for Claude, ChatGPT, Gemini, or other LLMs. Use when user mentions "optimize prompt", "improve prompt", "lyra", "prompt engineering", or needs help crafting effective AI prompts.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Converts natural-language descriptions or UI spec files into optimized Google Stitch prompts. Use when creating, refining, or validating design directives for Google Stitch. Use when user says "create a Stitch prompt", "optimize this for Stitch", "convert this spec to a Stitch prompt", "write a UI prompt", or mentions Google Stitch prompt authoring. Follows Stitch best practices with short, directive prompts focused on screens, structure, and visual hierarchy.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Use this when the user wants to generate, normalize, verify, deduplicate, or export training datasets for Codex, Antigravity, or Claude Code from topics, URLs, reference material, web research, or existing JSONL/CSV files. Supports SFT and DPO workflows, custom export schemas, and deterministic local pipeline scripts.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Evaluate code (especially LLM-generated) for structural simplicity using Rich Hickey's "Simple Made Easy" framework. Use this skill whenever reviewing a PR, diff, or code snippet for accidental complexity — particularly when the code was generated by an AI coding assistant and line-by-line review isn't feasible. Also use when the user asks about complecting, simplicity vs. easiness, structural coupling, or concept deduplication. Trigger on phrases like "is this simple", "does this complect", "review for complexity", "structural analysis", or any reference to Hickey, Simple Made Easy, or grey-box review.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Use when a request needs external sources, current facts, listings, a supplied URL, papers, or library and API documentation. When using Firecrawl for the request, use firecrawl_search to find sources and firecrawl_scrape to read a supplied or discovered URL. Firecrawl also retrieves documents, collects structured data, and tracks website changes. Respect explicit source and tool preferences.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Product Discovery through an 18-task methodology in 6 blocks — from market analysis to an investor-ready financial plan. Use when someone wants to run Product Discovery, validate a startup, assess a market hypothesis, prepare for an investment round, or raises an adjacent topic — idea validation, niche sizing, competitor teardown, product-market fit, pitch preparation, startup unit economics, geographic expansion into a new country, go-to-market for SEA / Asia / MENA. People rarely name PD explicitly; they say 'is there a market for this', 'assess my idea', 'I need a deck for an angel', 'we are entering Thailand/Singapore/UAE, how do we size it'. Use the skill in those cases too. Three modes: Light (~45 min, idea stage), Full (~2-3 h, MVP with customers), Geographic Expansion (~2 h, product entering a new geography). Produces 3-4 artifacts in /mnt/user-data/outputs/: one-pager.pptx, financial-plan.xlsx, presentation.pptx, optionally interview-guide.docx.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Use when creating high-quality scientific or medical infographic images or prompts, including review figures, disease maps, teaching diagrams, mechanism summaries, classification overviews, timelines, matrices, and academic visual explainers. If the user asks to generate the image, use ChatGPT image generation.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Use when creating scientific magazine cover style images or prompts, research hero visuals, journal-cover-inspired concepts, academic poster covers, PPT covers, and high-impact scientific visual concepts. If the user asks to generate the image, use ChatGPT image generation.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

ML Engineer — model deployment, MLOps pipeline, feature store, model monitoring, A/B testing, production ML system design

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Integrate LLM API RAG fine-tuning prompt engineering พร้อมคุม token cost และกัน prompt injection

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

เขียน SQL วิเคราะห์ insight สรุป finding และออกแบบ dashboard — แปลงตัวเลขเป็น action

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Check that the scheduled data work behind the reports actually ran and produced something sensible: the daily reshaping scripts, the API pulls, the tables the reports read from. Use each morning before anyone opens a report, when a report looks wrong and nobody knows whether it is the data or the query, after a schema or API change, when a new audience or source is added to the pipeline, or when a scheduled job failed or finished suspiciously fast.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Stand up reporting for a new audience — an external client, a business unit, a region, an executive team — end to end: what they need to see, whether the data exists, access and scoping, their place in the pipeline, the standard report set, and validation against numbers they can check themselves. Use when a new audience is onboarded, when an existing one adds a site, a region or a business unit, when their reporting is rebuilt after a change on their side, or when they say the numbers do not match their own records.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Turn a request for a report or a dashboard widget into something buildable: the question it answers, the metric and its definition, where the data comes from, the query shape, and what it will not answer. Use when a new report or dashboard is requested, when an existing one needs a metric added or changed, when a widget is asked for in a sentence and needs pinning down, or when two reports disagree about the same number.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Write the SQL or MongoDB query that answers a question and keeps answering it: the right grain, no fan out, parameterised window, checked against something known. Use when a number is needed that no report carries, when an existing query returns something that looks wrong, when a query is slow or expensive, or when somebody asks for a one off pull.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Fix data that has gone wrong and change shape without breaking what reads it: duplicates, nulls, inconsistent document shapes, a field that means two things, a column that has to change. Use when a report cannot be built because the data is not in the right shape, when duplicates, nulls or inconsistent shapes are producing wrong numbers, when a field is needed or has to change, before a migration, or when onboarding data from a new source that does not match the standard shape.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Comprehensive statistics expert covering probability theory, distributions, hypothesis testing, regression, and Bayesian methods

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Expert guidance across solar physics, planetary science, stellar evolution, cosmology, and observational techniques with 2025 mission data

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Expert in vector spaces, matrices, linear transformations, eigenvalues, and applications to data science and machine learning

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Use this skill whenever the user asks to run Stata commands, estimate econometric models, work with .dta files, run a .do file, generate Stata output, or do any statistical analysis where Stata is involved. Also trigger when the user mentions Stata variables, Stata syntax, or econometric tasks where Stata is the natural tool, including regressions, IV estimation, diff-in-diff, RDD, panel data, clustering, summary statistics, and margins. Stata runs through pystata on StataNow 19.5 BE; configure once with stata_setup, then drive everything with stata.run() and exchange data directly with pandas. Prefer this skill over subprocess calls or .do-file shelling for Stata work, including cases where the user does not say pystata.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Diagnose and fix slow pyfixest regression GRIDS (many feols/fepois calls run sequentially) that stay slow despite demeaner_backend="cupy64" and an idle GPU. Use when: (1) a script looping dozens of pf.feols models on a 100k+ row panel takes ~1 min/model, (2) process inspection shows ~1-1.5 cores busy and nvidia-smi shows ~0% GPU utilization with a resident cupy context, (3) planning any worker prompt that will run a model grid (robustness variants x FE structures x domains). Root cause: per-model CPU-side single-threaded fixed costs (formulaic model-matrix build, interaction construction, singleton detection, cluster vcov) dominate wall time; GPU demeaning is a small slice. Fix: shard the model grid across OS processes and/or use pyfixest multiple-estimation syntax; mandate this IN THE WORKER PROMPT.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Style every matplotlib figure like the Stata 18/19 default (stcolor) scheme — Arial embedded as TrueType, white background, recessive light-gray grid below the data, no top/right spines, unframed legends, and the validated blue/red/gray palette. Use whenever a task generates or restyles charts, plots, or figures for papers, reports, or slides, even if the user doesn't mention Stata — this is the house style for all publication figures. Also use when asked to make figures "look like Stata", match the stcolor scheme, or restyle existing matplotlib output.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Diagnose whether an LLM classifier's validation-gate failure is GOLD-BOUND before spending on prompt revision or model changes. Use when: (1) a scoring pipeline over-predicts a label (precision low, recall high) and a prompt clarification is proposed to tighten it, (2) a pilot/validation gate fails and the fix candidates are prompt edits, (3) inter-rater agreement on the weak label was already low (κ < ~0.6). Core check: if gold POSITIVES share the exact feature the revision would exclude, no prompt can pass a gold-scored gate — recall craters while precision barely moves. Also documents the verified surgical-pilot design (single-section diff, tune/holdout split, pre-registered gate, perturbation check on untouched sections).

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Raise real concurrency in asyncio LLM batch scorers built on the OpenAI SDK (AsyncOpenAI, including OpenAI-compatible providers like DeepSeek). Use when: (1) raising an asyncio.Semaphore above ~100 produces no throughput gain, (2) a batch pipeline saturates near 100 in-flight requests despite a larger semaphore, (3) planning a high-concurrency campaign against a provider with no hard rate limit (DeepSeek v4-flash tolerates 2000+ in flight). Root cause: AsyncOpenAI's default httpx pool caps max_connections at 100, silently bottlenecking any larger semaphore — you must pass a custom http_client with httpx.Limits sized to the semaphore.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.