Skills

Research (literature review)

Integrated investigation across the paper, patent, and web tracks — composes the active sub-skill tracks into a single [n]-cited report.

3 method skills composed: Research — papers · Research — patents · Research — web & industry

Research — papers

Academic-paper track — fans out across paper databases (PubMed·Crossref·OpenAlex·Semantic Scholar·Europe PMC·ERIC·arXiv, plus Scopus·WoS·KCI when connected).

Research — patents

Patent track — searches and aggregates patents via Google Patents (no key) plus PatentsView·KIPRIS (when connected), with filing-year, assignee, and status breakdowns.

Research — web & industry

Web & industry track — investigates company, institutional, and news sources via web search, with source-credibility distinctions.

Quick fact check (search → sourced one-line answer)

For a single factual question (who / when / where / how many / which version) — run one to three web or paper-database searches, read the most relevant page, and answer in one line with the source URL. No report, no clarifying survey, no literature pipeline; unverified answers are labelled as such.

Hypothesis discovery (gaps · contradictions · ranked hypotheses)

Evidence-based hypothesis discovery for a research problem — survey what prior studies did (tabulated), map the research space, detect gaps and contradictory findings, then propose 4–8 new testable hypotheses, each with its rationale, a novelty check against the closest study, the expected result and a study-design sketch, ranked by novelty, plausibility, testability and impact.

Experiment design

From a description of the study you want to run, survey how related studies designed their experiments, then produce a structured design report (goal · framework · participants · measures · constraints · the rationale and source behind every design choice) plus a protocol flow diagram, an expected-results chart, a power curve, and a summary of the reference studies' designs.

Data analysis (inspect → plan → analyze → interpret)

End-to-end analysis of an uploaded dataset — inspect structure and quality, write an explicit analysis plan, clean and transform with every step recorded, run the analyses with the verified statistical tools, and deliver a reproducible analysis report (data quality · plan · assumption checks · results with effect sizes and CIs · interpretation · limitations · code).

12 method skills composed: Descriptive statistics · Correlation analysis · Regression (linear · logistic) · Mixed-effects models · A/B testing · Speech-to-text (local whisper) · Acoustic features (formants · MFCC) · Voice emotion (acoustic features) · Facial expression (image · video, measured) · Body pose (image · video, measured) · Remote PPG from face video (rPPG) · EEG analysis (preprocess · band power · topomap · ERP · ERD/ERS · SSVEP)

Statistics (design → assumptions → method → diagnostics → effects)

Statistical method selection and testing for a dataset and hypothesis — identify the design, audit the assumptions, choose and justify the test or model, run it with the verified tools, run diagnostics, and report effects with uncertainty (effect size · CI · power) in a statistical report.

13 method skills composed: t-test (with assumptions) · One-way ANOVA · Power analysis · Bayesian estimation (Bayes factor) · Reliability (Cronbach α · McDonald ω) · Exploratory Factor Analysis · Confirmatory Factor Analysis · Structural Equation Modeling · Item Response Theory · Test selection & execution (parametric ↔ non-parametric) · Assumption checks & test selection · Effect size estimation · Multiple-comparison correction

Figure (diagram → editable vector figure)

Draw a scientific diagram as an editable native vector figure — study flow, protocol, conceptual framework, path/SEM model, analysis pipeline, concept tree, or a simple bar/line chart — from the Figure DSL, with journal templates, session citations on nodes, multi-panel composition, and SVG/PNG/PDF/PPTX export. Not for photographic or illustrative art (see Illustration).

Illustration (scientific vector illustration)

Generate a scientific illustration — laboratory glassware and equipment, a recording rig, a brain or body structure, a device, a person, or a whole apparatus scene — as an editable vector asset in the figure-tool style (semi-flat cel shading, thin grey outlines, muted palette, white background), or turn a procedure / methods paragraph into a journal-style figure in which every step is drawn and arranged as an arrowed flow (illustrated procedure). Image model → vectorised SVG with the background removed; points editable in the panel; placeable inside a diagram as a node icon. Off in sensitive projects and local-model mode.

Paper writing (claim–evidence mapping → manuscript)

Turn this project's validated research state — collected sources, analyses, figures, design documents — into a manuscript or manuscript sections (IMRaD + Conclusion), with every claim bound to a session citation [n] or to a session analysis result, an explicit claim–evidence map, and disclosure of anything that could not be supported.

Review (report review · peer review)

Umbrella for the two review tracks — report-review (independent methodological check of a document, analysis or design the user points at) and peer-review (journal-style peer review of a manuscript for a target journal). Picks the track from the request; /review runs whichever fits, /report-review and /peer-review pin one.

2 method skills composed: Report review (claims · methods · statistics · evidence · reproducibility) · Peer review (target journal · 5 reviewers · editorial decision)

Report review (claims · methods · statistics · evidence · reproducibility)

Independent scientific and methodological review of a research artifact the user points at — a report, analysis or study design (uploaded or created in this project), or a pinpointed check inside a document ("check the statistics in section 3"). Parses the claims, reviews methods and statistics, verifies cited evidence, checks reproducibility, and delivers a prioritized issue list with concrete fixes. A whole manuscript, thesis or dissertation handed over for review belongs to peer-review.

Peer review (target journal · 5 reviewers · editorial decision)

Journal-style peer review of a manuscript, thesis or dissertation draft the user attaches (PDF, DOCX, HWP or Markdown) — also the skill for "review my paper / 논문 초안 검토해줘" with no journal named — looks up the journal's aims & scope and submission rules, gathers the journal's recent comparable papers, runs deterministic checks (rules, reference registry, recomputed statistics), has five reviewer personas (journal fit · methodology · domain expert · broader perspectives · devil's advocate) score and comment, and a handling editor issue the decision (Accept / Minor Revision / Major Revision / Reject), a prioritized revision guide and recommended journals.

Study workflow (research → design → analysis → proposal)

End-to-end study pipeline in one turn — literature research, study design with independent review and one revision, statistical analysis plan with review, then an integrated study proposal (/study).

Insight (easy-to-read reports · facial emotion · heart rate · voice · posture · presentation rehearsal · review sentiment · ad reaction · A/B comparison · laugh count · brand colour · music mood · thumbnail)

Umbrella for plain-language reports made for non-researchers — marketers, event hosts, creators, consumers. Picks the report type from the request and the attached material and runs exactly one: insight-face (a face video or photo → an emotion report card), insight-heart (a face video → a heart-rate report card) insight-voice (a recording or a video's audio → a voice report card) insight-posture (a video or photo of a person's body → a posture report card) insight-presentation (one rehearsal video with sound → expression + voice + posture in one card) insight-reviews (reviews, comments or open answers as a file or pasted text → a sentiment report card), insight-ad (an ad video + viewers' reaction videos → reaction by scene), insight-compare (two creatives or two viewer groups → side-by-side differences), insight-laugh (a face video or photo → a fun laugh count card), insight-color (logo, packaging or thumbnail images → a brand colour card), insight-music (a music file or a video's sound → a mood card) or insight-thumbnail (thumbnails, posters or packaging → a first-impression card). /insight runs whichever fits, /insight-face, /insight-heart, /insight-voice, /insight-posture, /insight-presentation, /insight-reviews, /insight-ad, /insight-compare, /insight-laugh, /insight-color, /insight-music and /insight-thumbnail pin one.

12 method skills composed: Facial emotion report (face video · photo) · Heart-rate report (face video) · Voice report (recording · video audio) · Posture report (presentation · filming video · photo) · Presentation rehearsal report (one video · expression + voice + posture) · Review sentiment report (reviews · comments · open answers) · Ad reaction report (ad video + viewers' reaction videos) · A/B reaction comparison (two creatives or two viewer groups) · Laugh count card (face video · photo) · Brand colour card (logo · packaging · thumbnails) · Music mood card (BGM · jingle · video sound track) · Thumbnail first-impression card (thumbnails · posters · packaging)

Facial emotion report (face video · photo)

Turns an uploaded face video or photo into an easy-to-read emotion report card for non-experts — the mix of basic emotions (with neutral), four key numbers, the emotion flow over time, highlight moments with face thumbnails and the reaction to each scene; with several people in view, every person is analysed and compared. Made for marketing and events (ad or content reactions, pop-ups, unboxing, creator clips), not for clinical, hiring or personality assessment.

Heart-rate report (face video)

Turns an uploaded face video (15 s or longer, face front-on in steady light) into an easy-to-read heart-rate report card for non-experts — average, highest and lowest heart rate, the flow over time against the first 10 seconds, the moments the heart beat fastest, rose fastest and was calmest, and the reaction to each scene. Heart rate is estimated from skin-colour changes on the face (camera rPPG), no wearable needed. Made for marketing and events (ad, trailer, game or content reactions, pop-ups, presentation practice), not for medical, fitness or health assessment.

Voice report (recording · video audio)

Turns an uploaded recording (or the audio track of a video, 10 s or longer, one person speaking) into an easy-to-read voice report card for non-experts — speaking pace, pitch variation, vocal energy, pauses, the flow over time, the strongest moment, highest pitch and longest pause with what was said then, voice steadiness and the numbers for each part. Made for creators, presenters and marketers (narration, podcasts, live commerce, pitch or interview practice, customer calls), not for emotion detection, speaker assessment or clinical voice evaluation.

Posture report (presentation · filming video · photo)

Turns an uploaded video or photo of a person whose upper body is visible into an easy-to-read posture report card for non-experts — body steadiness, gesture activity, shoulder tilt and how much of the time the body faced the camera, the flow over time, the biggest gesture, the most sway and the longest steady stretch with skeleton thumbnails, where the hands stayed and the numbers for each part; a photo gets shoulder, head and body tilt with an overlay. Made for presentation, pitch or interview rehearsal, filming, live commerce and event clips, not for medical, rehabilitation or fitness assessment.

Presentation rehearsal report (one video · expression + voice + posture)

Turns ONE rehearsal video (a presentation, pitch, lecture or interview practice, 20 s to 5 min, filmed from the front with sound) into an easy-to-read report card that combines facial expression, voice and posture — bright expressions and facing the camera, speaking pace and pauses, body steadiness and gestures, a combined flow over time, the strongest moment, the longest pause and time looking away with face thumbnails, the numbers for the opening, main part and closing, and up to three "what to watch next time" observations. Made for students, job seekers, lecturers and creators rehearsing on camera, not for grading, hiring or personality assessment.

Review sentiment report (reviews · comments · open answers)

Turns customer reviews, app-store reviews, social-media comments or open survey answers — an uploaded CSV/XLSX/TSV/TXT or text pasted into the chat (5 to 500 items) — into an easy-to-read sentiment report card for non-experts — the positive / neutral / negative mix (AI-classified, with a re-check agreement), the top complaint topic, sentiment by topic (price, quality, delivery, design, service, usability), frequent words, six representative sentences, the rating distribution and a monthly trend when dates exist. Made for marketers, shop owners and creators, not for statistical research or identifying individual writers.

Ad reaction report (ad video + viewers' reaction videos)

Turns an ad, trailer or content video plus 1–6 videos of viewers' faces watching it into an easy-to-read report card — the ad's scenes as a thumbnail strip with reaction strength, the viewers' average positive expressions, the strongest and quietest scenes, a reaction flow aligned to the ad, a scene-by-scene table (dominant expression, agreement between viewers, optional heart-rate change), a viewer summary by number and highlight moments pairing an ad frame with a viewer's face. Reaction clips are aligned by the ad's sound heard in them, by given start times, or assumed to start together. Made for marketers and agencies testing creatives, not for measuring ad effectiveness or purchase intent.

A/B reaction comparison (two creatives or two viewer groups)

Compares two groups of viewer reaction videos — the same ad watched by group A and group B, or two different ads each with its own viewers — in one easy-to-read card: side-by-side numbers (positive expressions, expressiveness, strongest scene, heart-rate rise, viewers), the two reaction flows overlaid, the difference by scene in percentage points (same ad) or each group's strongest scenes (different ads). Differences only, no winner, no statistics; a group under 3 viewers is flagged as a rough guide. Made for marketers and agencies comparing creatives, not for research.

Laugh count card (face video · photo)

Turns a face video or photo into a fun, shareable laugh count card — how many times the person laughed, for how long, the longest laugh, laughs per minute, how often the eyes smiled too, a laugh timeline and three moments with face thumbnails (biggest smile, longest laugh, first laugh); several people are numbered and their laughing-together time is shown; a photo gets a smile score. Made for selfies, reaction clips, watch parties and events — for fun, never for judging or typing anyone.

Brand colour card (logo · packaging · thumbnails)

Turns 1–5 images (a logo, packaging, thumbnails, feed posts) into an easy-to-read brand colour card — a six-colour palette with shares, HEX codes and plain colour names, tone tiles (brightness, saturation, warmth, contrast), WCAG contrast checks for text on the main colours and on white, colour-vision previews and, with several images, how consistent the palettes are. All computed by code. Made for founders, small brands and creators without a designer, not a brand guideline.

Music mood card (BGM · jingle · video sound track)

Turns a music file or a video's sound track (5 s to 5 min) into an easy-to-read mood card — tempo in BPM, energy and brightness, the estimated key, an energy flow with quiet, medium and loud sections, the most intense, calmest and biggest build-up moments, two mood axes (calm–intense, dark–bright) and one or two mood words marked as an estimate. Made for people choosing background music for ads, shorts and events, not for genre, quality or copyright judgement.

Thumbnail first-impression card (thumbnails · posters · packaging)

Turns 1–4 images (YouTube thumbnails, posters, packaging fronts, the top of a product page) into an easy-to-read first-impression card — an attention heat-map with the three spots the eye is drawn to first, attention focus, face share, clutter, colour pop and contrast, how the image looks shrunk to phone and desktop sizes, rule-of-thirds placement and a comparison table for several images. Attention is estimated from image statistics, not measured on real viewers. Made for creators and shop owners, not a click-through prediction.

Compact conversation

Summarize the conversation so far into a compact, faithful summary and continue from it — for long chats where earlier turns crowd the context (/compact).