{
  "markdown": "# research-co-pilot\n\n[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://github.com/Marazii/research-co-pilot/blob/main/LICENSE)\n[![Version](https://img.shields.io/badge/version-0.11.4-blue.svg)](https://github.com/Marazii/research-co-pilot/blob/main/CHANGELOG.md)\n[![CI](https://github.com/Marazii/research-co-pilot/actions/workflows/validate.yml/badge.svg)](https://github.com/Marazii/research-co-pilot/actions/workflows/validate.yml)\n[![Claude Code](https://img.shields.io/badge/Claude%20Code-plugin-8b5cf6.svg)](#installation--claude-code)\n[![Claude.ai](https://img.shields.io/badge/Claude.ai-skills-d97757.svg)](#installation--claudeai)\n[![Cite](https://img.shields.io/badge/cite-CITATION.cff-green.svg)](./CITATION.cff)\n[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.20186491.svg)](https://doi.org/10.5281/zenodo.20186491)\n\n<!--\n  The DOI badge resolves to the Zenodo concept DOI for this repo, which\n  always points to the latest release. Version-specific DOI for v0.10.0 is\n  10.5281/zenodo.20186492 (in CITATION.cff identifiers). On future releases,\n  the update-citation-doi.yml workflow auto-updates both — see\n  CONTRIBUTING.md § Release process.\n-->\n\n\nA Claude co-pilot for the entire research lifecycle — from picking a question through final peer review. Designed for academic researchers (graduate students, postdocs, faculty) and applied researchers (UX, policy, public health, data science) who want a rigorous collaborator that respects methodological standards instead of generating plausible-looking output.\n\nWorks in two places:\n\n- **Claude Code** (CLI / IDE) — installs as a plugin with skills, slash commands, and subagents.\n- **Claude.ai** (web / desktop) — install individual skills as Claude Skills.\n\nSame skills, same behavior, two surfaces.\n\n---\n\n## What it helps with\n\n| Stage of research | Skill | Trigger examples |\n|---|---|---|\n| Finding a question | [**research-brainstorm**](./skills/research-brainstorm/README.md) | \"help me find a thesis topic in X\", \"brainstorm research ideas\" |\n| Reviewing the field | [**literature-review**](./skills/literature-review/README.md) | \"what does the research say about X\", \"do a lit review on Y\" |\n| Designing the study | [**methodology-advisor**](./skills/methodology-advisor/README.md) | \"should I use an RCT or quasi-experiment?\", \"what sample size do I need?\" |\n| Stress-testing the ethics | [**ethics-committee**](./skills/ethics-committee/README.md) | \"is this study ethical?\", \"review my IRB application\", \"draft an ethics statement\" |\n| Designing instruments | [**survey-design**](./skills/survey-design/README.md) | \"is this question biased?\", \"find a validated scale for X\" |\n| Analyzing quantitative data | [**data-analysis**](./skills/data-analysis/README.md) | \"clean this dataset\", \"fit a regression\", \"run a power analysis\" |\n| Analyzing qualitative data | [**qualitative-coding**](./skills/qualitative-coding/README.md) | \"code these transcripts\", \"build a codebook\", \"find themes\" |\n| Drafting the manuscript | [**manuscript-drafter**](./skills/manuscript-drafter/README.md) | \"write the intro\", \"draft the discussion\", \"write the abstract\" |\n| Replicating someone else's study | [**replication-designer**](./skills/replication-designer/README.md) | \"design a replication of X\", \"is this finding robust?\" |\n| Funding the work | [**grant-writer**](./skills/grant-writer/README.md) | \"draft my specific aims\", \"NSF broader impacts\", \"ERC synopsis\" |\n| Presenting it | [**talk-builder**](./skills/talk-builder/README.md) | \"turn this paper into a 12-min conference talk\", \"job talk outline\", \"thesis defense slides\" |\n| Responding to reviewers | [**reviewer-response**](./skills/reviewer-response/README.md) | \"draft my R1 response\", \"rebuttal letter\", \"address reviewer comments\" |\n| Formatting references | [**citation-formatter**](./skills/citation-formatter/README.md) | \"format these in APA\", \"fix my bibliography\" |\n| Getting feedback | [**peer-review**](./skills/peer-review/README.md) | \"review my paper\", \"fact-check this draft\", \"review my slides\" |\n\nEvery skill is grounded in research methods literature and refuses common AI failure modes — no fabricated citations, no p-hacking, no glossed-over disagreement between sources, no qualitative \"themes\" without an audit trail, no embellished findings, no voice-mismatched drafts.\n\nEach skill name above links to its **per-skill README** with a synthetic example, the trigger phrases, composition with other skills, and honest caveats. The [`examples/`](./examples/) folder contains a worked sample for each shipped skill.\n\n---\n\n## How the skills work together\n\nThe skills aren't 14 islands — they're an **orchestrated network**: skills are nodes, shared artifacts are edges, and `/research` is the conductor. Each skill reads what upstream skills produced, does its one job, writes a predictable artifact, and points at what comes next.\n\n```\nresearch-brainstorm\n  └→ literature-review\n       └→ methodology-advisor ──→ ethics-committee        (audit before data)\n            ├→ survey-design          (if an instrument is needed)\n            └→ [DATA COLLECTION — your turn]\n                 ├→ data-analysis ──→ stats-validator (independent second look)\n                 └→ qualitative-coding\n                      └→ manuscript-drafter ──→ citation-formatter\n                           ├→ peer-review            (pre-submission audit)\n                           │    └→ reviewer-response  (after an R&R)\n                           └→ talk-builder           (conference talk)\n\ngrant-writer          ← draws on brainstorm + methodology (funding; parallel track)\nreplication-designer  ← draws on a target paper; can seed a fresh cycle\n```\n\nTwo ways to use the network:\n\n- **Pipeline mode** — say `/research start a project on <topic>` and the conductor walks the whole lifecycle, invoking each skill in turn (Claude Code), reading and writing the project vault (a clean, templated `research/<project>/` folder), and **pausing at the steps only you can do** (collecting data, getting IRB approval, submitting). It resumes where you left off.\n- **Pairwise handoffs** — run any single skill; when it needs an upstream artifact or finishes with an obvious next step, it *offers* to chain to the relevant skill (Claude Code) or tells you which `/command` to run next (claude.ai). It never auto-runs another skill without your go-ahead.\n\nSkills coordinate through **the project vault** — a clean, visible, templated folder (`research/<project>/`) where every skill files its output into a numbered lifecycle-stage subfolder (`01-ideation/`, `02-literature/`, … `10-dissemination/`), alongside a `knowledge/` folder and an auto-generated project `README.md`. No more hunting for where a file went; the project reads top-to-bottom like its own story. Vault use is opt-in — run any skill standalone and it behaves exactly as before.\n\n```\nresearch/<project>/\n  README.md  manifest.json   knowledge/   00-inbox/\n  01-ideation/  02-literature/  03-methodology/  04-ethics/  05-instruments/\n  06-data/      07-analysis/    08-drafts/        09-review/  10-dissemination/\n  audits/   archive/\n```\n\nBeyond filing the documents, the vault holds the project's *knowledge*: canonical facts (sample N, IRB#, pre-registration, target journal, language), one shared bibliography every skill cites from, the decisions log, consolidated open questions, the glossary, the voice profile, and entities (pseudonyms only — never PII). Each skill reads the vault at intake (so it never re-asks your sample size or re-fabricates a citation) and updates it at output. `/vault audit` scans every document against the vault to **catch drift before a reviewer does** — the abstract that says N=240 while the methods say 247, a citation that doesn't resolve, a term used three ways, PII in the wrong file. `/vault organize` keeps the folder tidy. Full contract: [`docs/research-vault.md`](./docs/research-vault.md).\n\nFull network map, manifest schema, the human-gate rule, and the Claude Code ↔ claude.ai degradation table: [`docs/skill-network.md`](./docs/skill-network.md).\n\n---\n\n## Why use this instead of plain Claude\n\nPlain Claude is great at fluency, weak at rigor. Researchers feel this most when:\n\n- It invents DOIs and authors that don't exist.\n- It runs statistical tests without checking assumptions.\n- It generates \"themes\" from qualitative data without showing where they came from.\n- It writes literature reviews that average away real disagreement in the field.\n- It says \"more research is needed\" instead of identifying specific gaps.\n- It drafts manuscript prose that doesn't match the author's existing voice.\n\nEach skill in this plugin is a structured workflow that catches these failures: source verification before citation, assumption diagnostics before reporting effects, codebook + memos for every theme, surfacing of conflicting findings, specific rather than generic gaps, mandatory voice extraction before drafting. The skills are written in the voice of a methodologist who has supervised dissertations, reviewed for journals, and published in mixed methods.\n\nFor a longer discussion of the design philosophy, see [`docs/philosophy.md`](./docs/philosophy.md). For common questions about safety, citation, language coverage, and how this compares to SaaS research-AI products, see [`docs/faq.md`](./docs/faq.md).\n\n---\n\n## Installation — Claude Code\n\nYou need Claude Code installed and a working terminal.\n\n**One-command install** (recommended):\n\n```bash\n# Add this repo as a Claude Code marketplace\nclaude plugin marketplace add Marazii/research-co-pilot\n\n# Install the plugin from that marketplace\nclaude plugin install research-co-pilot@research-co-pilot-marketplace\n```\n\nAfter install, run `/research` for the menu, or invoke any specific skill by name (e.g. `/lit-review climate adaptation in coastal cities`).\n\n**Updating an existing install:**\n\n```bash\nclaude plugin marketplace update research-co-pilot-marketplace\nclaude plugin update research-co-pilot@research-co-pilot-marketplace\n```\n\nThen restart Claude Code.\n\n**Local install** (for development or to use without GitHub):\n\n```bash\ngit clone https://github.com/Marazii/research-co-pilot.git\ncd research-co-pilot\n\n# Add your local clone as a marketplace\n# Edit ~/.claude/settings.json and add:\n{\n  \"plugins\": {\n    \"marketplaces\": {\n      \"research-co-pilot-local\": {\n        \"type\": \"local\",\n        \"path\": \"/absolute/path/to/the/parent/of/research-co-pilot\"\n      }\n    }\n  }\n}\n\n# Then in Claude Code:\nclaude plugin install research-co-pilot@research-co-pilot-local\n```\n\n---\n\n## Installation — Claude.ai\n\nEach skill is uploaded to claude.ai individually as a Claude Skill. Pre-built upload bundles ship as **GitHub Release assets** — stable URLs that don't change when commits land on `main`.\n\n1. Go to the [**latest release**](https://github.com/Marazii/research-co-pilot/releases/latest) and download the `.zip` for each skill you want.\n\n   Direct downloads always pointing at the latest stable release:\n\n   ```\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/literature-review.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/methodology-advisor.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/ethics-committee.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/data-analysis.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/qualitative-coding.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/research-brainstorm.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/manuscript-drafter.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/replication-designer.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/grant-writer.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/talk-builder.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/reviewer-response.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/citation-formatter.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/survey-design.zip\n   https://github.com/Marazii/research-co-pilot/releases/latest/download/peer-review.zip\n   ```\n\n2. In claude.ai, open **Settings → Capabilities → Skills** (Pro/Team/Enterprise plans), or open your workspace's **Skills library** (Team/Enterprise).\n\n3. Click **Upload skill** and select a `.zip`. Repeat per skill.\n\nOnce uploaded, claude.ai loads the skill automatically when your message matches its trigger phrases (e.g. \"do a literature review on…\" loads `literature-review`).\n\n> Subagents and slash commands are Claude Code-only and don't apply in claude.ai. The skills handle their work directly when run there.\n\n---\n\n## Visibility & assets\n\nScreenshots, demo GIFs, and the social-preview image live under [`docs/`](./docs/):\n\n- [`docs/philosophy.md`](./docs/philosophy.md) — design choices and what the plugin won't do.\n- [`docs/faq.md`](./docs/faq.md) — common questions about citations, regulated data, language coverage, comparison to SaaS products.\n- [`docs/screenshots/`](./docs/screenshots/) — high-DPI screenshots of typical outputs (annotated PDFs, methodology with the AI/ML extensions table, etc.).\n- [`docs/demo.gif`](./docs/demo.gif) — ~30-second screencast of a typical flow.\n\n*(Some visibility assets are added during the v0.10.0-rc.1 → v0.10.0 final cycle and may not all be present yet.)*\n\n---\n\n## Usage examples\n\n### \"I'm a year into my PhD and I have no idea what to study.\"\n\n```\n/brainstorm I work on misinformation and trust in news, broadly. I have access to\npanel survey data and can run online experiments. Constraint: must be defensible\nin a quant-focused department.\n```\n\nThe skill generates 15-25 candidate questions across descriptive, causal, predictive, and contrarian framings; scores them on interest / answerability / novelty / feasibility; sharpens the top 3 into full study sketches with predicted findings, risks, and follow-up studies. See [`examples/research-brainstorm/`](./examples/research-brainstorm/) for a worked sample.\n\n### \"I need to write a literature review for my thesis.\"\n\n```\n/lit-review the effect of remote work on early-career mentorship in knowledge industries\n```\n\nThe skill searches, reads, fact-checks, and synthesizes — organizing by idea (not by source), surfacing where studies disagree, marking each claim with a confidence tag, producing a complete review with appraisal table. See [`examples/literature-review/`](./examples/literature-review/) for a worked sample.\n\n### \"Reviewer 2 said my analysis is wrong.\"\n\n```\n/analyze ./data/study2_clean.csv — they want me to handle the clustered structure\n(students within classrooms within schools). Currently I just have OLS.\n```\n\nThe skill walks through the data, fits the appropriate mixed-effects model, runs diagnostics, compares specifications, and writes up the result with effect sizes, CIs, and an honest discussion of what changes from the original. See [`examples/data-analysis/`](./examples/data-analysis/) for a worked sample.\n\n### \"I have 18 interview transcripts and need themes for a paper.\"\n\n```\n/code-themes ./transcripts/\n```\n\nThe skill anonymizes, develops a codebook (inductively, deductively, or hybrid — you choose), codes systematically, runs inter-rater reliability if you have a second coder, develops themes with disconfirming-case checks, and writes the qualitative findings section with quoted excerpts and a reflexivity statement. See [`examples/qualitative-coding/`](./examples/qualitative-coding/) for a worked sample.\n\n### \"I got an R&R and need to draft the response letter.\"\n\n```\n/respond review_round1.md manuscript_v1.md  (R1 from J. Hypothetical Studies)\n```\n\nThe skill categorizes every reviewer point (concession / partial / pushback / clarification / out-of-scope / minor), drafts the response per point with the reviewer's verbatim quote, drafts the corresponding manuscript revisions (delegating long-form prose to `manuscript-drafter` to preserve voice), assembles a one-page cover letter that surfaces points of disagreement up front, and self-audits that every claimed revision actually appears in the revised manuscript.\n\n### \"Extend my Discussion chapter — match the rest of the thesis.\"\n\n```\n/draft Discussion section, target ~1500 words.\nInputs: existing_draft.md, methodology.md, qual_findings.md, references.bib.\nLanguage: Hebrew.\n```\n\nThe skill extracts a voice profile from the existing manuscript (sentence length, person/voice, hedge intensity, signature phrases, connectors, citation style, Hebrew register markers), drafts new prose to match, enforces a per-section word budget, scans for non-academic register, and grounds every new idea in your bibliography (or flags it as `[LITERATURE NEEDED]` with concrete search guidance). See [`examples/manuscript-drafter/`](./examples/manuscript-drafter/) for a worked sample showing the v0.8.0 overhaul.\n\n---\n\n## What's in the box\n\n```\nresearch-co-pilot/\n├── .claude-plugin/\n│   ├── plugin.json           # Claude Code plugin manifest\n│   └── marketplace.json      # one-command install from GitHub\n├── .github/\n│   ├── workflows/            # CI: validate / build-dist / release\n│   ├── ISSUE_TEMPLATE/       # bug / feature / new-skill-proposal\n│   └── PULL_REQUEST_TEMPLATE.md\n├── skills/                   # Portable: work in Claude Code AND claude.ai\n│   ├── literature-review/\n│   ├── methodology-advisor/\n│   ├── ethics-committee/\n│   ├── data-analysis/\n│   ├── qualitative-coding/\n│   ├── research-brainstorm/\n│   ├── manuscript-drafter/\n│   ├── replication-designer/\n│   ├── grant-writer/\n│   ├── talk-builder/\n│   ├── reviewer-response/\n│   ├── citation-formatter/\n│   ├── survey-design/\n│   ├── peer-review/\n│   └── vault/                # the project's knowledge librarian (facts, bibliography, audit)\n│       └── (each skill has SKILL.md + README.md + optional reference files)\n├── agents/                   # Subagents (Claude Code only)\n│   ├── source-finder.md      # parallel reading of many academic sources\n│   ├── data-cruncher.md      # heavy compute in isolation\n│   ├── transcript-coder.md   # bulk qualitative processing\n│   ├── manuscript-drafter.md # long-form drafting in isolation\n│   ├── stats-validator.md    # independent second-look on a colleague's analysis\n│   └── vault-organizer.md    # read a pile of files + file them into the vault's categories\n├── commands/                 # Slash commands (Claude Code only)\n│   ├── research.md           # /research — entry-point router\n│   └── (one per skill: /lit-review, /methodology, /ethics, /analyze,\n│        /code-themes, /brainstorm, /draft, /replicate, /grant, /talk,\n│        /respond, /cite, /survey, /peer-review, /vault)\n├── examples/                 # Synthetic minimal example per skill\n├── docs/                     # skill-network, research-vault, philosophy, faq, screenshots, demo\n├── scripts/\n│   └── build-zips.sh         # rebuild dist/ after editing skills\n├── CHANGELOG.md\n├── CITATION.cff              # academic citation metadata\n├── CONTRIBUTING.md\n├── CODE_OF_CONDUCT.md\n├── SECURITY.md\n├── LICENSE                   # MIT\n└── README.md\n```\n\n### Skills (work in both surfaces)\n\n| Skill | What it does |\n|-------|-------------|\n| [`literature-review`](./skills/literature-review/README.md) | Fact-checked synthesis of a body of work. Verifies citations, organizes by idea, surfaces disagreement, marks confidence on every claim, identifies specific gaps. Supports narrative, systematic, scoping, rapid, and thematic reviews. |\n| [`methodology-advisor`](./skills/methodology-advisor/README.md) | Quant + qual research design — picking the right design for the question, sampling strategy, power and sample size, validity threats, IRB/ethics, pre-registration. Includes mandatory creative AI / ML / Big Data extensions section that forces the researcher to consider non-conventional methods. |\n| [`ethics-committee`](./skills/ethics-committee/README.md) | Simulates an IRB / REC / HREC pre-submission review. Audits informed consent, risk-benefit, vulnerable populations, data privacy, deception, payment, AI/LLM use, social media data, equity in recruitment. Produces a decision letter with required revisions. Optional 3-reviewer panel mode. Not a substitute for institutional approval. |\n| [`data-analysis`](./skills/data-analysis/README.md) | End-to-end quantitative work in Python or R: cleaning, EDA, statistical testing, modeling (regression, mixed-effects, predictive, time series, survival), assumption diagnostics, sensitivity analyses, visualization, reproducible scripts. |\n| [`qualitative-coding`](./skills/qualitative-coding/README.md) | Codebook development and application using thematic analysis, grounded theory, IPA, framework analysis, or content analysis. Inter-rater reliability (Cohen's κ, Krippendorff's α). NLP-assisted exploration for large corpora — with required validation against hand-coding. |\n| [`research-brainstorm`](./skills/research-brainstorm/README.md) | Generates 15-25 research ideas via question-form variations, cross-field grafts, and contrarian moves. Scores them. Sharpens the top 3 into study sketches. Pushes past the obvious next study. |\n| [`manuscript-drafter`](./skills/manuscript-drafter/README.md) | Drafts long-form manuscript sections from your methodology and analysis outputs. **Preserves the existing manuscript's voice** via mandatory voice-profile extraction. **Enforces hard per-section word budgets** with a compression pass. **Enforces academic register** via a banned-pattern audit. **Grounds every new idea in literature** (distinguishes `[CITATION NEEDED]` from `[LITERATURE NEEDED]`). Two-pass ideation → prose workflow. Drafts in the manuscript's language (English, Hebrew, etc.). |\n| [`replication-designer`](./skills/replication-designer/README.md) | Designs direct, conceptual, generalization, or robustness replications. Extracts the original spec, justifies every deviation, computes adequate replication power, plans pre-registration, pre-specifies replication-success criteria, supports multi-site logistics. |\n| [`grant-writer`](./skills/grant-writer/README.md) | Drafts proposal sections (Specific Aims, Significance, Innovation, Approach, Broader Impacts, DMP, lay summary, biosketch, budget justification) tuned to NIH (R01/R21/F31/K), NSF (Standard / CAREER / GRFP), ERC, Wellcome, Horizon Europe, or foundation grants. Won't overpromise. Honest about scheme fit. |\n| [`talk-builder`](./skills/talk-builder/README.md) | Turns one or more papers 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 length, audience, format, and discipline conventions. Deck-platform-agnostic outline plus optional Marp / Quarto / reveal.js / Beamer stubs. |\n| [`reviewer-response`](./skills/reviewer-response/README.md) | Drafts rigorous, polite, point-by-point response to reviewer comments (R1 / R2 / R3) plus matching manuscript revisions. Categorizes each comment, drafts response + revision, assembles cover letter. Won't concede a point the data doesn't support. Composes naturally with `peer-review` and `manuscript-drafter`. |\n| [`citation-formatter`](./skills/citation-formatter/README.md) | APA 7, MLA 9, Chicago (NB and AD), Harvard, Vancouver, IEEE, AMA, journal-specific. Verifies DOIs, handles edge cases (preprints, datasets, software, AI tools). Generates BibTeX/RIS. Document-wide consistency check. |\n| [`survey-design`](./skills/survey-design/README.md) | Question wording, scale choice, ordering effects, response burden. Recommends validated instruments rather than inventing new ones. Includes a pilot plan (cognitive interviews + quantitative pilot). Translation guidance. |\n| [`peer-review`](./skills/peer-review/README.md) | Multi-mode rigorous review: paper verdict, homework grading, committee panel, fact-check audit, plagiarism check, draft thinking-partner, presentation feedback, post-review iterate. **Returns a reviewed file with annotations anchored at the relevant locations** across `.docx` (inline comments + tracked changes), `.pdf` (sticky-note comments + highlights via PyMuPDF), `.pptx` (native PowerPoint comments on slides / shapes), `.tex` (`% REVIEWER:` line comments, optional `changes` package). |\n| [`vault`](./skills/vault/README.md) | The project's knowledge librarian (infrastructure, not a lifecycle stage). Manages the **research vault** — canonical facts (sample N, IRB#, pre-reg, journal, language), one shared bibliography, the decisions log, consolidated open questions, glossary, voice profile, and entities (pseudonyms only — never PII). `/vault audit` scans every document against the vault to catch cross-artifact drift (mismatched sample sizes, unresolved citations, term inconsistency, accidental PII). Files stay the source of truth; the vault is reconciled. |\n\n### Subagents (Claude Code only — claude.ai has no equivalent)\n\n| Agent | Use it when |\n|-------|-------------|\n| `source-finder` | Reading >5 academic sources in parallel without bloating the main conversation. Returns structured digests with verified citations. |\n| `data-cruncher` | Heavy computation, many model variants, simulations, sensitivity grids — in isolation. Returns a tight results summary instead of raw output. |\n| `transcript-coder` | Bulk cleaning, anonymization, and code-application across many transcripts. Returns coded JSON + a summary instead of full transcript text. |\n| `manuscript-drafter` | Long-form drafting (whole sections or whole papers) in isolation so the parent doesn't get flooded with thousands of words. Returns a structured digest plus the draft as files. |\n| `stats-validator` | Independent second-look on a colleague's analysis. Reads their script + data + report in fresh context (no narrative contamination), re-runs, sensitivity-checks, and returns a tight memo with confidence judgment. |\n| `vault-organizer` | Reading through a pile of files (a messy folder, a Downloads dump, an inherited project) and filing them into the vault's categories. Classifies each file → stage folder with confidence + rationale, flags PII (never files real-name keys / signed consent / raw identifiers into the vault), and returns a filing plan; on approval, files non-destructively (copy by default). Spawned by `/vault organize <path>`. |\n\n### Slash commands (Claude Code only)\n\n`/research` is the entry point if you're not sure which skill you need — it routes by description, or conducts the whole pipeline. The other commands invoke the matching skill directly: `/lit-review`, `/methodology`, `/ethics`, `/analyze`, `/code-themes`, `/brainstorm`, `/draft`, `/grant`, `/replicate`, `/talk`, `/respond`, `/cite`, `/survey`, `/peer-review`, and `/vault` (project knowledge + drift audit).\n\n---\n\n## Output conventions\n\n- **Claude Code:** primary deliverable written to a markdown file in your working directory (e.g. `lit_review_<topic>.md`, `analysis_<topic>.md`, `methodology_<study>.md`). Reproducible scripts saved alongside.\n- **Claude.ai:** primary deliverable rendered as a downloadable artifact; code and data files saved in the analysis sandbox for download.\n\nEither way, the deliverables are designed to be portable — markdown + scripts you can drop into Overleaf, Google Docs, R Markdown, Quarto, or your repo.\n\n---\n\n## Customizing\n\nEach skill's `SKILL.md` is the single source of truth for its behavior. Edit it to:\n\n- Change trigger phrases (frontmatter `description`).\n- Add discipline-specific guidance (e.g. CONSORT for clinical trials, COREQ for qualitative).\n- Adjust the output format to match your lab's or journal's expectations.\n\nAfter editing skills, rebuild the claude.ai bundles:\n\n```bash\n./scripts/build-zips.sh\n```\n\nThen re-upload the changed `dist/*.zip` files to claude.ai (or push to a tag and CI builds + uploads to the Release automatically).\n\n---\n\n## Contributing\n\nContributions are welcome with a high bar. Start with [`CONTRIBUTING.md`](./CONTRIBUTING.md) — it documents:\n\n- What the project is (and is not).\n- The PR-level bar (validation, CHANGELOG, version bump, dist rebuild, per-skill README + example updates).\n- The path for new skills (open a `New skill proposal` issue first; PRs without prior discussion are declined).\n- What gets declined and why.\n\nBy contributing you agree to the [Code of Conduct](./CODE_OF_CONDUCT.md) (Contributor Covenant 2.1). Security issues should follow the [SECURITY.md](./SECURITY.md) disclosure process.\n\n---\n\n## Citing\n\nIf you use this plugin in your research, please cite it. The repository's [`CITATION.cff`](./CITATION.cff) renders as a \"Cite this repository\" widget on GitHub, producing APA, BibTeX, and other formats. A Zenodo-minted DOI will be available from the v0.10.0 final release onward.\n\n---\n\n## Philosophy\n\nFor a longer write-up of the design choices — why each skill encodes specific hard rules, what \"rigor over fluency\" means in practice, and what the plugin won't do — see [`docs/philosophy.md`](./docs/philosophy.md).\n\n---\n\n## Caveats and honest limits\n\n- **Citations:** the literature-review skill verifies what it can, but published research changes. Treat the output as a strong first pass that you read and confirm — never paste it into a manuscript without checking.\n- **Statistics:** the data-analysis skill picks reasonable defaults and checks assumptions, but applied stats often involves judgment calls a model can't make. Treat its output as a draft analysis your statistician reviews.\n- **Qualitative coding:** AI-assisted coding is exploratory. Use it to surface candidates, then validate against hand-coded subsets before relying on labels at scale.\n- **Peer review:** useful for catching blind spots and surface issues. Cannot replace expert review in your specific subfield.\n- **Ethics and IRB:** the methodology and ethics-committee skills flag IRB-relevant considerations but are not a substitute for your institution's review process.\n- **Voice preservation:** the manuscript-drafter and reviewer-response skills extract voice from your existing prose. Without an existing manuscript, voice cannot be enforced.\n\nIn short: this is a force multiplier for a researcher who knows what good work looks like — not a replacement for that judgment.\n\n---\n\n## License\n\nMIT — see [LICENSE](LICENSE).\n\n---\n\n## Credits\n\nBuilt by Maya Arazi — all skills (including `peer-review`), subagents, and slash commands. Methodological frameworks reference standard research methods literature (Braun & Clarke, Charmaz, Shadish-Cook-Campbell, Dillman, Lincoln & Guba, and many others) — all citations belong to their original authors.\n",
  "bytes": 30834,
  "sha": "512f51bc5ae62d913da79a552c873117c9ff2d933ee8e9e532eadfbcdf82f9b4",
  "repo_slug": "marazii/research-co-pilot",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_marazii_research_co_pilot_research_co_pi_40a0c311/readme"
}