Skip to content

kgraph57/paper-writer-skill

paper-writer

Verified

Medical/scientific paper writing workflow skill. Manages the full pipeline from literature search to submission-ready manuscript. Creates and manages a project directory with IMRAD-format section files, literature matrix, reference management, and quality checklists. Supports both English and Japanese papers. Triggers: 'write paper', 'paper-write', 'start manuscript', '論文を書く', '論文執筆', '論文プロジェクト', 'manuscript', 'research paper', '原稿作成'.

FileKindSize
SKILL.mdskill md77.7 KB
examples/case-studies/discovery-to-draft.mdmarkdown5.3 KB
references/abstract-formats.mdmarkdown5.1 KB
references/adversarial-review.mdmarkdown21.6 KB
references/ai-disclosure.mdmarkdown6.4 KB
references/ai-for-science-model.mdmarkdown12.8 KB
references/citation-guide.mdmarkdown2.4 KB
references/citation-verification.mdmarkdown10.5 KB
references/clinical-trial-registration.mdmarkdown4.7 KB
references/coauthor-review.mdmarkdown14.2 KB
references/coi-detailed.mdmarkdown6.8 KB
references/desk-rejection-prevention.mdmarkdown6.1 KB
references/hook-compatibility.mdmarkdown6.1 KB
references/humanizer-academic.mdmarkdown12.0 KB
references/imrad-guide.mdmarkdown2.5 KB
references/journal-reformatting.mdmarkdown5.6 KB
references/journal-selection.mdmarkdown13.5 KB
references/keywords-guide.mdmarkdown4.6 KB
references/master-reference-list.mdmarkdown11.3 KB
references/multilingual-guide.mdmarkdown12.4 KB
references/novelty-check.mdmarkdown13.8 KB
references/open-access-guide.mdmarkdown4.6 KB
references/pubmed-query-builder.mdmarkdown16.7 KB
references/reporting-guidelines-full.mdmarkdown10.6 KB
references/reporting-guidelines.mdmarkdown3.2 KB
references/section-checklist.mdmarkdown4.5 KB
references/statistical-reporting-full.mdmarkdown8.6 KB
references/statistical-reporting.mdmarkdown22.5 KB
references/submission-portals.mdmarkdown4.1 KB
references/supplementary-materials.mdmarkdown3.9 KB
references/tables-figures-guide.mdmarkdown7.7 KB
references/word-count-limits.mdmarkdown4.6 KB
scripts/analysis-template.pyexecutablescript19.6 KB
scripts/compile-manuscript.shexecutablescript6.9 KB
scripts/forest-plot.pyexecutablescript16.9 KB
scripts/sr-dedup.pyexecutablescript12.5 KB
scripts/sr-pdf-link.pyexecutablescript6.8 KB
scripts/sr-prisma-count.pyexecutablescript8.7 KB
scripts/table1.pyexecutablescript9.7 KB
scripts/word-count.shexecutablescript12.8 KB
templates/abstract.mdmarkdown1.3 KB
templates/acknowledgments.mdmarkdown1.7 KB
templates/analysis-workflow.mdmarkdown5.1 KB
templates/case-abstract.mdmarkdown1.2 KB
templates/case-introduction.mdmarkdown1.7 KB
templates/case-report.mdmarkdown2.2 KB
templates/conclusion.mdmarkdown1.0 KB
templates/cover-letter.mdmarkdown1.3 KB
templates/data-management.mdmarkdown5.7 KB
templates/declarations.mdmarkdown6.1 KB
templates/discussion.mdmarkdown1.8 KB
templates/graphical-abstract.mdmarkdown2.7 KB
templates/highlights.mdmarkdown3.8 KB
templates/human-loop-ledger.mdmarkdown9.4 KB
templates/introduction.mdmarkdown1.8 KB
templates/limitations-guide.mdmarkdown6.2 KB
templates/literature-matrix.mdmarkdown1.2 KB
templates/methods.mdmarkdown1.9 KB
templates/preregistration.mdmarkdown19.5 KB
templates/project-init-case.mdmarkdown3.2 KB
templates/project-init.mdmarkdown3.6 KB
templates/proof-correction.mdmarkdown2.5 KB
templates/research-question.mdmarkdown18.6 KB
templates/response-to-reviewers.mdmarkdown12.5 KB
templates/results.mdmarkdown1.5 KB
templates/review-article.mdmarkdown11.6 KB
templates/revision-cover-letter.mdmarkdown10.3 KB
templates/sr-data-extraction.mdmarkdown9.2 KB
templates/sr-grade.mdmarkdown14.7 KB
templates/sr-outline.mdmarkdown13.5 KB
templates/sr-prisma-flow.mdmarkdown12.9 KB
templates/sr-prospero.mdmarkdown15.9 KB
templates/sr-rob.mdmarkdown18.3 KB
templates/sr-screening-pipeline.mdmarkdown12.1 KB
templates/study-design.mdmarkdown28.5 KB
templates/submission-ready.mdmarkdown2.5 KB
templates/title-page.mdmarkdown3.5 KB

references/ai-for-science-model.md

View on GitHub
# The AI-for-Science Operating Model

> The rest of this skill is a *manuscript factory*: it assumes the research
> question and the data already exist, and helps you express, format, and submit
> them. This document is the *research engine* that sits upstream and around it.
> It turns the linear pipeline into a loop and names the two things only a human
> can supply.

---

## 1. Why this document exists

Three frontier systems (2024–2026) defined where "AI for Science" actually is,
and — more usefully — where it isn't:

| System | What the AI did | What stayed human | Lesson |
|--------|-----------------|-------------------|--------|
| **Sakana AI Scientist v2** (2025) | Hypothesis → experiment → analysis → full manuscript, fully autonomous; one paper passed an ICLR *workshop* review (6/7/6) | **Nobody.** No human in the loop | External review found hallucinations, faked results, novelty inflation, and a misattributed citation (LSTM). Authors admitted it would not pass a *conference* bar. **Full autonomy without a human anchor produces confident slop.** |
| **Google AI co-scientist** (Nature, 2026) | Multi-agent *generate → debate → rank → evolve* of hypotheses, grounded in literature + databases | **Wet-lab validation, clinical judgment** | Reproduced an AMR hypothesis in days that took a lab years. But Google states plainly: it "did not achieve autonomous discovery, did not complete clinical trials, did not replace the research team." **The value is the hypothesis engine; truth is decided by the world.** |
| **FutureHouse Robin** (2025) | Hypotheses, experiment design, data analysis, manuscript figures — all AI | **The physical experiments** | Discovered + validated a novel therapeutic candidate; concept-to-submission in 2.5 months. The AI owned the *intellectual scaffold*; humans owned *contact with reality*. |

The pattern is identical across all three. **AI compresses the cost of everything
between idea and reality toward zero. The human's irreducible contribution narrows
to exactly two things: the idea, and the data.** Everything else — search,
synthesis, design drafting, statistics, prose, formatting, compliance, revision —
is execution that AI now does at full power.

This skill is built for a **clinician publishing under their own name**, where a
fabricated finding is not an embarrassment but a patient-safety and
career-integrity event. So we take the frontier's *capability* and bolt on the
*discipline* the frontier systems lacked.

---

## 2. The two human-sovereign inputs

Mark these two gates in every project. They are the only points where the human
is not replaceable — protect them, and automate aggressively everywhere else.

### 💡 IDEA — judgment, meaning, ethics

The human owns:
- **What is worth asking** — clinical relevance, not just statistical tractability.
- **What the finding *means*** — the bridge from a coefficient to a child at a bedside.
- **What is ethical** — consent, equipoise, who is studied, who benefits.
- **Whether to publish at all** — the "is this true and useful enough to put my name on it" call.

AI *proposes* ideas (and should, abundantly — see `templates/research-question.md`).
The human *selects, kills, and reframes* them. An AI-generated hypothesis that no
human chose to pursue is noise, not science.

### 📊 DATA — contact with reality

The human owns:
- **The ground truth** — real patients, real outcomes, the愛育 outpatient clinic. AI cannot generate this; it can only generate *plausible-looking* versions of it, which is the failure mode.
- **The provenance** — IRB approval, de-identification, the chain of custody from chart to CSV.
- **The integrity of measurement** — what was actually measured, with what error.

AI *analyzes, models, and visualizes* data. It must **never originate a data point,
a participant, or a result.** This is the single hardest rule in the skill
(`Section-Specific AI Guidelines` in SKILL.md), and AI-for-Science makes it *more*
load-bearing, not less.

> **The thesis, stated once:** As the marginal cost of execution falls to zero, a
> scientist's job becomes *only* the two things a machine cannot do — supply real
> data and exercise judgment about ideas. Build the workflow so that those two are
> the explicit, protected pivots, and let the machine do the rest at maximum power.

---

## 3. The loop (not the line)

The base skill is a line: `Search → Outline → Draft → Submit`. The research engine
wraps that line in a discovery loop:

```
        💡 IDEA gate (human selects)
                │
   ┌────────────▼─────────────┐
   │  Phase −1: DISCOVERY      │   ← NEW
   │  observation → questions  │   templates/research-question.md
   │  → novelty check          │   references/novelty-check.md
   │  → study design + power   │   templates/study-design.md
   │  → PRE-REGISTRATION lock  │   templates/preregistration.md  ← anti-HARKing gate
   └────────────┬─────────────┘
                │  (plan is now frozen)
        📊 DATA gate (human supplies real data; IRB verified)
                │
   ┌────────────▼─────────────┐
   │  Phases 0–3: EXECUTE      │   existing skill
   │  analysis → draft (IMRAD) │   (now bound to the frozen plan)
   └────────────┬─────────────┘
                │
   ┌────────────▼─────────────┐
   │  Phase 6.5: RED-TEAM      │   ← NEW
   │  try to KILL the claim    │   references/adversarial-review.md
   │  before a reviewer does   │   agent: paper-red-team
   └────────────┬─────────────┘
                │ survives?
        ┌───────┴────────┐
       YES              NO → back to Discovery (the finding was an artifact)
        │
   Phases 4–10: humanize → submit → revise
```

The loop can run more than once. A red-team kill is not a failure — it is the
system working. Sakana v2's papers shipped *because nothing tried to kill them.*

---

## 4. Division of labor across the full loop

| Step | AI does | Human owns | Guardrail |
|------|---------|-----------|-----------|
| Observe | — | **💡 the clinical observation** | — |
| Generate questions | proposes 5–15 candidates, scores FINER | **selects / kills / reframes** | novelty check (§5.2) |
| Design study | drafts design, comparator, power, analysis plan, DAG | **approves; owns ethics & feasibility** | pre-registration (§5.1) |
| Pre-register | drafts the registration record | **submits it; it is now binding** | the lock itself |
| Collect data | — | **📊 supplies real, IRB-approved data** | never AI-originated (§2) |
| Analyze | runs the *pre-registered* analysis, then clearly-labeled exploratory | verifies against raw data | analysis = frozen plan first |
| Draft | writes all IMRAD prose | verifies every number traces to data | base skill Phase 3 |
| Red-team | attacks its own main claim from 4 angles | adjudicates the verdict | adversarial review (§5.3) |
| Submit / revise | drafts everything | signs, decides, corresponds | base skill Phases 7–10 |

Rule of thumb: **if a step can be checked against an external truth (a database, a
raw data file, a statistical assumption), AI may execute it. If a step *defines*
what counts as true or worthwhile (the question, the data, the meaning), a human
must own it.**

---

## 5. The three integrity guardrails

These are what make an AI-accelerated study *more* rigorous than a hand-built one,
instead of less. Each one prevents a documented frontier failure mode.

### 5.1 Pre-registration — prevents HARKing & the garden of forking paths

The danger of a fast loop is that AI can hypothesize-after-results-are-known
(HARKing) and p-hack across a multiverse of analyses at machine speed, dressing
post-hoc noise as an a-priori finding. The only defense is to **freeze the
hypotheses and the primary analysis plan before the data gate** (OSF / UMIN-CTR /
jRCT / PROSPERO). After the lock:
- Pre-registered analyses are reported first, exactly as planned.
- Everything else is labeled **exploratory** and cannot carry a confirmatory claim.
- Deviations from the plan are disclosed, not hidden.

See `templates/preregistration.md`. For systematic reviews this already exists as
the PROSPERO step — this generalizes it to primary studies.

### 5.2 Novelty check — prevents reinvention & novelty inflation

Sakana v2 *overestimated its own novelty*. Before committing to a question, run an
automated "has this already been answered?" sweep against real literature
(PubMed MCP, OpenAlex, Europe PMC, Semantic Scholar) and classify the gap honestly:
genuinely novel / incremental / already-answered / contested. A question that the
literature already settles is killed at near-zero cost — the cheapest experiment
is the one you don't run. See `references/novelty-check.md`.

### 5.3 Adversarial self-review — prevents confident slop

The difference between Robin (credible) and Sakana v2 (slop) was that nothing in
Sakana's loop *tried to make the paper fail.* Before submission, a hostile internal
panel attacks the main claim from four angles — statistical, methodological,
novelty, integrity — and a "steelman the null" pass argues that the finding is
chance, confounding, or bias. If the claim cannot survive its own red team, it does
not go to a journal's red team. See `references/adversarial-review.md` and the
`paper-red-team` agent.

---

## 6. The autonomy dial

How much to automate is a *setting*, not a fixed property. Choose per project, and
state it in the README.

| Mode | AI runs… | Human touches… | Appropriate for |
|------|----------|----------------|-----------------|
| **Manual** | one step at a time, asks before each | everything | learning the method; high-stakes flagship paper |
| **Co-pilot** (default) | a full phase, stops at every gate | both sovereign gates + every gate verdict | most clinical research |
| **Autopilot** | the whole loop between gates, reports after | only the 💡 and 📊 gates + final sign-off | low-stakes, well-trodden designs; never the first time |

**Hard rule for medical / clinical work:** the 💡 IDEA gate, the 📊 DATA gate, and
the pre-registration lock are **never** autopilot. No setting removes the human
from "is this ethical," "is this data real," and "is this what we committed to
test." This is non-negotiable and overrides any "team mode" or stage-gate
automation elsewhere in the skill.

---

## 7. How this maps onto the existing skill

- **New upstream:** Phase −1 (Discovery) runs *before* Phase 0 (Project Init). Its
  output — a frozen, pre-registered, novelty-checked, powered study plan — becomes
  the input the rest of the skill always assumed you already had.
- **Re-enterable:** Phase −1 is not a strict linear chain with one fixed start. An
  advanced study (question or draft protocol already in hand, but pre-data) enters at
  the *first guardrail it hasn't yet passed* — typically novelty, then a design audit,
  prereg, and a design-stage red-team — rather than re-forging a question it already
  has. The loop has no single fixed entry point (see the SKILL.md Phase −1 entry matrix).
- **New self-critique:** Phase 6.5 (Adversarial Review) runs after Quality Review
  (Phase 6) and before Pre-Submission (Phase 7).
- **Upgraded plumbing:** Phase 1 literature search and Phase 5 citation verification
  switch from "WebSearch is unreliable, ask the user" to real literature APIs
  (PubMed MCP et al.).
- **New instrumentation:** `templates/human-loop-ledger.md` records, at every gate,
  whether the decision was 💡 human-judgment, 📊 human-data, or 🤖 AI-execution —
  making the two sovereign inputs auditable.
- **Unchanged:** everything downstream of the data gate. The manuscript factory is
  already excellent; the research engine just feeds it better inputs and stress-tests
  its outputs.

---

## 8. One-paragraph summary for the README

> This skill operates the full scientific loop, not just the writing of it. AI
> generates and ranks research questions, checks their novelty against the live
> literature, designs the study and its power, and — after a human supplies real
> data and locks a pre-registered plan — runs the analysis, drafts the manuscript,
> and red-teams its own main claim before any journal sees it. Two things stay
> human and protected at all times: the **ideas** (what is worth asking, what it
> means, what is ethical) and the **data** (real, IRB-approved, never
> machine-originated). Everything between is executed at full power, under
> pre-registration, novelty, and adversarial-review guardrails that make the science
> more rigorous, not less.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall