`/neuroflow:paper`¶

Draft and internally review every section of your manuscript — nothing is saved without critic approval.

/paper is the unified manuscript command. It replaces two separate steps (draft, then review) with a single brutal loop: the paper-writer agent drafts each section, the paper-critic agent applies the full eight-area peer-review methodology, and the loop iterates up to 3 times per section. Only approved sections reach disk.

When to use it¶

After /data-analyze — results and figures are ready
You want to produce a manuscript that has been through rigorous internal review before it reaches a real journal
You want drafting and critique in one integrated workflow with automatic critic approval gates

What it does¶

Claude reads your project memory and asks:

Target journal?
Format? LaTeX or Markdown/Word
Which section(s)? — or full paper
Review focus? — full eight-area critique on every section, or specific areas

For each section, the paper-writer agent first returns an outline. You approve it (or change it) before any prose is written. The agents run as subagents and never ask you anything themselves: every question comes from the main conversation.

The writer cites only papers in your project library (.neuroflow/ideation/papers/ or your reference-manager export). A claim that needs a paper the library lacks gets a [CITE: …] placeholder and a question to you; it is never filled from memory.

The write→critique loop¶

Every section goes through this loop before it is saved:

paper-writer → draft v1
paper-critic → [STATUS: APPROVED] or [STATUS: REJECTED] + actionable feedback

if APPROVED → section saved to manuscript/
if REJECTED → paper-writer revises → draft v2
              paper-critic → verdict

if APPROVED → section saved
if REJECTED → paper-writer revises → draft v3
              paper-critic → verdict

if APPROVED → section saved
if REJECTED (3rd) → loop halts; unresolved critique logged to .neuroflow/paper/critic-log.md
                    user decides whether to accept the draft or stop

Maximum 3 iterations per section. Nothing is written to manuscript/ without [STATUS: APPROVED] or explicit user acceptance. Rounds 2 and 3 continue the same writer and critic agents, so they keep what they already read and the critic remembers its own earlier feedback.

Modes¶

Mode	What it does
default	The drafting loop above
`--submit`	Submission package: cover letter, availability statements, CRediT, suggested reviewers, AI-use statement, acknowledgements line, pre-submission checks (citations, statistics, figures) and a ✅/❌ checklist
`--revise`	Reviewer rebuttal under the prime rule: minimal changes, only what a reviewer asked; an audit script checks that every change traces to a comment
`--coauthor <file.docx>`	Works through a coauthor's Word comments and tracked changes one row at a time; the coauthor's file is never edited
`--abstract`	Conference abstract with one critic pass
`--xray <file>`	Sentence-by-sentence check: an annotated copy with red (countable defects) and orange (judgement) findings; reports only. With the neuroflow mod, `--xray view` opens the findings as a pane, red first, eight to a page: Enter on accept or reject, then your yes, decides one, and `j` and `k` step through them; `--xray check <file>` runs only the statistics and citation scripts
`--auto on\\|off\\|status\\|sync`	A living paper skeleton in `.neuroflow/paper/`: facts with sources, preregistered hypotheses quoted word for word, one Results slot per planned test, reporting gaps — no prose results. With the neuroflow mod, `--auto status` opens a pane with the gaps and the sources changed since the last sync; Enter on sync now there, after your yes, runs the sync

Checks that run as scripts¶

Script	What it reports
`cite_check.py`	"DOI resolves" or "does not resolve"; retraction, withdrawal, expression-of-concern and correction notices in Crossref ("no retraction notice found in Crossref as of "); cited DOIs missing from the project library
`statcheck.py`	Whether each reported p matches its t, F, r, χ² or z and degrees of freedom; corrected p-values are skipped
`revise_audit.py`	Changes in the revised manuscript that no reviewer comment quotes, and quoted changes that are not in the manuscript
`docx_comments.py`	A .docx's comments and tracked changes as a work table

They live in the phase-paper skill (scripts/), need only Python 3.10+, and never claim more than they test: a DOI that resolves can still be cited for something the paper does not say.

AI use and style editing¶

The humanizer skill is a style edit (filler, rhythm, register) that runs only when you ask for it. It is never used to hide AI involvement. --submit drafts an AI-use statement from the session and critic logs; you confirm it before it goes into the package.

Critic standards¶

The paper-critic agent applies the full neuroflow:review-neuro eight-area methodology to every draft:

Area	What is checked
Language & Terminology	Spelling, grammar, neuroscience terminology errors, causality language
Internal Consistency	Figure/table references, value consistency across sections
Claim Support & Causality	Overclaims, causality creep, FC over-interpretation
Statistics	Effect sizes, multiple-comparison correction, null models, estimator specification
Methods Reproducibility	COBIDAS/ARRIVE compliance, artefact thresholds, data/code availability
Contribution & Novelty	Novelty claims, prior work comparison, journal fit
Literature Gap	Missing citations vs local paper library and the manuscript's own references
Figure Review	Figure citations, panel completeness, axes/units/error bars, statistical annotations

Figure Review findings are advisory: they never block a verdict, and you see them in the critique summary after each section — fix a figure in its plotting code through the /data-analyze figure check. A figure that contradicts the text or the statistics is a blocking Internal Consistency or Statistics item instead.

A section is approved only if it would survive peer review at a top-tier neuroscience journal. The bar is not "acceptable draft" — it is "ready for submission".

Drafting order¶

For a full paper, sections are drafted in this order:

Order	Section	Source material
1	Methods	`.neuroflow/experiment/`, `.neuroflow/data-preprocess/`, `.neuroflow/data-analyze/`
2	Results	`.neuroflow/data-analyze/` (analysis outputs, figures)
3	Introduction	`.neuroflow/ideation/` (research question, literature)
4	Discussion	All prior sections plus interpretation
5	Abstract	Always last — summarises after all other sections are approved

Output¶

Approved sections are saved to manuscript/ (or the path set in .neuroflow/paper/flow.md). Loop state is tracked in .neuroflow/paper/critic-log.md.

Files read and written¶

Direction	Files
Reads	`.neuroflow/project_config.md`, `.neuroflow/flow.md`, `.neuroflow/ideation/flow.md`, `.neuroflow/ideation/papers/`, `.neuroflow/data-analyze/flow.md`, `.neuroflow/preregistration/flow.md`, `.neuroflow/paper/flow.md`, `.neuroflow/sessions/` (`--submit`)
Writes	`.neuroflow/paper/` (critic log, coauthor rounds, X-ray files, skeleton, DOI cache), `.neuroflow/paper/flow.md`, `.neuroflow/reasoning/paper.jsonl`, `.neuroflow/project_config.md` (`target_journal`, `paper_auto`), `.neuroflow/timeline.md`, `.neuroflow/sessions/YYYY-MM-DD.md`, `manuscript/` (approved drafts, `submission/`, `revision/`, `abstracts/`)

/data-analyze — generate the results that go into the paper
/review — peer review a colleague's paper using the same eight-area methodology
/poster — the conference poster for the same results
/output — archive data and code before --submit claims availability

/neuroflow:paper¶