Marazii

citation-formatter

Format citations and bibliographies in any major academic style — APA 7, MLA 9, Chicago (notes-bibliography and author-date), Harvard, Vancouver, IEEE, AMA, and discipline-specific journal styles. Converts between formats, builds reference lists from raw input, validates DOIs, and generates BibTeX/RIS exports. Trigger when: user asks to "format this citation", "convert to APA", "bibliography in Chicago", "reference list", "BibTeX", "fix my references", "is this APA correct", or runs /cite.

data-analysis

End-to-end quantitative data work — cleaning, exploration, statistical testing, modeling, visualization, and reproducible scripting in Python or R. Handles messy CSVs, survey data, time series, panel data, and small-to-medium datasets. Produces analysis scripts and a written interpretation. Trigger when: user mentions "clean this data", "analyze", "statistics", "regression", "EDA", "explore the data", "fit a model", "visualize", "summary stats", "hypothesis test", "ANOVA", "t-test", "correlation", "Python script", "R script", "pandas", "tidyverse", or runs /analyze.

ethics-committee

Act as a research ethics committee — stress-test a protocol the way an IRB / REC / HREC would. Reviews informed consent, risk-benefit balance, vulnerable populations, data privacy, deception, debriefing, payment, dual-use risks, AI/LLM use in research, and equity in recruitment. Produces a committee-style decision letter with required revisions, recommended changes, and approval pathway. Useful before IRB submission, when responding to IRB feedback, when drafting ethics sections of a paper or grant, or when a protocol changes mid-study. Trigger when: user mentions "ethics review", "IRB", "REC", "HREC", "ethics committee", "ethics statement", "informed consent", "is this study ethical", "vulnerable population", "deception in research", "Belmont", "Helsinki", "GDPR research", "ethics approval", "ethics application", or runs /ethics.

grant-writer

Draft sections of a grant proposal — specific aims / lay summary / significance / innovation / approach / broader impacts / budget justification / data management plan / biosketch — adapted to the funder's format and review criteria. Supports NSF, NIH (R01, R21, R03, F31, F32, K-series), ERC (Starting, Consolidator, Advanced), Wellcome, Horizon Europe, NSF GRFP, foundation grants, and other major schemes. Works from research-brainstorm and methodology-advisor outputs, plus the funder's published criteria. Trigger when: user mentions "grant proposal", "specific aims", "NSF", "NIH", "ERC", "Wellcome", "Horizon Europe", "research proposal", "fellowship application", "biosketch", "broader impacts", "data management plan", "lay summary", "case for support", or runs /grant.

literature-review

Conduct rigorous, fact-checked academic literature reviews. Synthesizes sources, traces citation chains, flags weak claims, and produces structured outputs (narrative, systematic, scoping, or thematic). Trigger when: user asks for a "literature review", "lit review", "background research", "state of the field", "what does the research say about…", "summarize the literature on…", "find sources on…", or runs /lit-review. Works from sources the user provides (PDFs, links, citations) AND from web/database search when allowed.

Data & Documents Listed

manuscript-drafter

Draft long-form manuscript sections — abstract, introduction, related work, methods, results, discussion, limitations, conclusion — from a methodology document and an analysis report. Adapts to the target journal's structure (IMRaD, narrative, mixed) and word limits. Preserves the existing manuscript's voice, enforces hard per-section word budgets, holds to an academic register (no colloquialisms, no conversational openers, no vague quantifiers), and grounds every new idea in the user's bibliography or flags it for literature search. Drafts in the language of the existing manuscript (English, Hebrew, etc.). Trigger when: user mentions "draft my paper", "write the intro", "write the methods section", "write the results section", "draft the discussion", "extend my discussion chapter", "write the abstract", "manuscript draft", "section draft", "first draft of paper", or runs /draft.

methodology-advisor

Advise on quantitative and qualitative research methodology — design, sampling, validity, reliability, measurement, ethics, analysis plan, and — mandatory in every methodology output — creative AI / ML / Big Data extensions tailored to the specific research question. The creative-extensions section is not optional: it forces researchers to consider non-conventional, modern methods before settling on a design. Trigger when: user asks about "study design", "research design", "methodology", "what method should I use", "sample size", "power analysis", "sampling strategy", "validity", "reliability", "IRB", "pre-registration", "RCT vs quasi-experiment", "qualitative vs quantitative", "AI methods for my study", "ML approach", "big data approach", "creative methods", or runs /methodology.

qualitative-coding

Code qualitative data — interview transcripts, open-ended survey responses, field notes, documents, social media — using inductive, deductive, or hybrid approaches. Generates and refines codebooks, applies codes consistently, identifies themes, and supports inter-rater reliability checks. Includes NLP-assisted techniques (sentiment, topic modeling, embeddings) for exploring large corpora. Trigger when: user mentions "thematic analysis", "code transcripts", "qualitative coding", "codebook", "themes", "open coding", "axial coding", "grounded theory", "content analysis", "inter-rater", "Cohen's kappa", "topic modeling", "NLP on transcripts", or runs /code-themes.

replication-designer

Design a direct, conceptual, or generalization replication of a published study. Walks through identifying the target effect, extracting the original design with enough fidelity to replicate it, deciding what to hold equivalent vs. what to update, calculating sample size for adequate replication power (typically ~2.5x the original under reasonable assumptions), pre-registering the replication on OSF / AsPredicted, multi-site logistics if applicable, and statistically assessing replication success (significance + effect size + meta-analytic synthesis with the original). Trigger when: user mentions "replicate this study", "replication design", "registered replication", "many-labs", "is this finding robust", "direct replication", "conceptual replication", "preregistered replication", "replication power", or runs /replicate.

research-brainstorm

Generate, sharpen, and pressure-test research ideas — research questions, hypotheses, study angles, novel contributions, and contrarian framings. Pushes beyond obvious next steps to find what's actually worth studying. Useful at the start of a project or when stuck. Trigger when: user asks to "brainstorm research", "generate research ideas", "research questions", "hypothesis ideas", "what should I study", "I'm stuck on what to research", "thesis topic ideas", "novel angles", "what's interesting about", or runs /brainstorm.

reviewer-response

Draft a rigorous, polite, point-by-point response to reviewer comments on a journal or conference submission (R1 / R2 / R3 letters), plus the corresponding manuscript revisions. Categorizes each comment as concession / pushback / clarification needed / minor fix, drafts the response and the revision for each, and assembles a cover letter. Composes with `peer-review` (consume comments from a review output) and `manuscript-drafter` (revise prose using the manuscript's voice profile and academic register rules). Won't concede a point the data doesn't support; preserves framing where the reviewer misread; addresses every point. Trigger when: user mentions "respond to reviewers", "R1 response", "R2 response", "rebuttal", "address reviewer comments", "reviewer-response letter", "point-by-point response", "revision and resubmission", "response to reviewer 2", "cover letter to editor", or runs /respond.

survey-design

Design rigorous surveys and questionnaires — question wording, response scales, ordering, validation, pilot testing, and translation. Spots leading questions, double-barreled items, social desirability traps, and acquiescence bias. Recommends validated instruments where they exist instead of inventing new ones. Trigger when: user says "design a survey", "survey questions", "questionnaire", "Likert scale", "question wording", "is this question biased", "validated scale for", "survey pilot", or runs /survey.

talk-builder

Turn a paper (or several, a thesis, a body of work) into an academic talk — outline, per-slide content, speaker notes, opening hook, single take-home message, backup slides for Q&A, rehearsal plan. Adapts to talk length (3-min lightning through 90-min defense), audience (specialists / general field / cross-disciplinary / clinical / public), format (contributed / lightning / invited / plenary / keynote / symposium / workshop / defense / job talk / public lecture / course lecture), and discipline conventions (sciences, social sciences, CS / ML / HCI, medicine, education, humanities, law, math). Produces a deck-platform-agnostic outline plus optional Marp / Quarto / reveal.js / Beamer stubs. Trigger when: user mentions "turn paper into talk", "presentation outline", "slides for my talk", "conference talk", "lecture outline", "academic presentation", "speaker notes", "thesis defense talk", "job talk", "keynote", "invited talk", "lightning talk", "elevator pitch of my paper", or runs /talk.

vault

Manage the research project's vault — the clean, visible, templated project folder (research/<project>/) where every file lives in a numbered lifecycle-stage place, plus the knowledge layer (canonical facts like sample size / IRB# / journal, one shared bibliography, decisions log, open questions, glossary, voice profile, entities — pseudonyms only, never PII). Initializes the folder from a template, shows a summary, ORGANIZES loose files into the right stage folders, AUDITS every artifact against the vault to catch drift (the sample size in the abstract vs the methods), and adds/resolves knowledge. Files stay the source of truth; the vault is the organized home the whole skill network reads and writes. Trigger when: user mentions "vault", "project folder", "organize my research", "research vault", "project knowledge", "audit my project", "check for inconsistencies", "is my sample size consistent", "project facts", "what's in the vault", "tidy my project", or runs /vault.