/write

One entry point for writing every section of a paper. The researcher says what to write; this skill loads the procedure for that section and runs it. The sections are not interchangeable, so each has its own model, interview and checks in references/.

Usage

/write abstract
/write results-discussion
/write introduction src/introduction.tex
/write cover-letter                      # asks for the journal if unknown
/write                                   # infers the section, else asks

Sections

Methods is not offered. The course proposes no model for it, so none was invented; see todo/ in the repository. If asked, say so and offer to help in plain conversation instead.

Before anything else

1. Check for paper-out/STATE.md and read it, with decisions.md and the relevant interviews/*.md, before asking anything. State what you already know in two or three lines.
2. Load the style rules. Invoke paper-writing-style and read its references/human-voice.md and references/latex-conventions.md. The project's own CLAUDE.md outranks them.
3. Lock the language. Language of the user's first message, whole run. LaTeX output is always English.

Pick the section

1. From the argument. Match it loosely: results, discussion and results and discussion are one section; conclusion, rebuttal, reviewer response, cover, letter all map to a row above.
2. From a file path, if given: the file's name or contents say which section it is.
3. From context, if there is no argument: the open file, or the next step in paper-out/STATE.md. Say what you inferred and let the researcher correct it.
4. Otherwise ask. Offer the sections most likely given the state (at most four, plus "other"), or list all seven in chat. Never guess a section and start writing.

An unrecognised section is a question, not an error: show the table and ask.

Pick the mode

Look for the section in src/ and in main.tex, following the \input list rather than assuming file names.

• No draft exists. Run the reference's Create mode.
• A draft exists. Say what you found and ask which applies: update it (the reference's Update mode offers reconcile with the paper, apply suggestions, or resume), or rewrite it from scratch (Create mode, after confirming that the existing text may be discarded). If the researcher's request already says "update", "revise", "fix" or "finish", skip the question.
• Cover letter and response to reviewers often live outside the manuscript. Ask for the file if you cannot find it, and treat a letter written for another journal as a draft to retarget.

Run it

Read the chosen reference in full and follow it. Do not reimplement it here and do not mix procedures from two sections.

After it finishes

• Update paper-out/ before you finish, as references/session-state.md in paper-writing-style sets out, creating it if absent: this section's row in STATE.md, interviews/<section>.md, a decisions.md entry for each decision, open-questions.md for anything unresolved. Name the files you wrote. A run is not finished until this is done.
• Check the ripple: results changed means abstract, conclusions and the introduction gap may be stale; conclusions changed means the abstract's closing and the title may be stale. Name the affected sections and suggest /write <section> for each. Do not run them unasked.
• Suggest /review <section> on what was just written, and /ask-paper for the next step.

Hard rules

• Never invent a result, number, method detail or citation key. Emit the project's \todo{} macro, or % TODO: where there is none.
• Never write to the researcher's files before they approve the text. paper-out/ is the exception: update it every run, and say so.
• Never emit an em dash (---).
• Never start writing until the section is settled and the reference's interview has what it needs.
• Never write a section the course gives no model for. Say so instead.
• Never finish a run without updating paper-out/. Say which files you changed.