agentleFS
AlexyskoutnevAlexyskoutnev/research-skills/method-writing/SKILL.md

method-writing

Skill2 starsChanged 3 months agoMITView on GitHub

Write or revise a paper's Method section using recipes distilled from dissecting top ML researchers' papers. Use when drafting or restructuring a method section, choosing its architecture figure, algorithm box, or notation, deciding how to introduce and defend a system's components, or fixing a method that reads as a flat wall of mechanics. Applies the N-named-parts skeleton, the setup-banks-notation rule, the star-component discipline, the running-example rule, and mechanical checklists.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Follow method-guide.md in this folder, the complete plain-English method.

    line 8
  2. Distinctive:

    Algorithm box with defaults: If there is a real procedure or loop, box it: titled with the system name, line-numbered, SMALL-CAPS …

    line 30
  3. Distinctive:

    Seal the section: No benchmark numbers, no results plot, no standalone related work, no limitations paragraph, those live in other sections.

    line 34

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Method section (master method)

Follow method-guide.md in this folder, the complete plain-English method. This file is the operating summary.

Process

  1. Skeleton first. Name the N parts (the count gravitates to three) in one roadmap sentence, in the exact order you will install them, no math in the opener. Then install exactly those N parts, one at a time, until the system visibly assembles. No part may appear that the roadmap did not promise.
  2. Setup subsection banks notation. A dedicated first subsection defines every symbol at first use under bold run-in headers and fences off borrowed formalism (what you did not invent but need). Nothing floats; no symbol appears before its definition.
  3. One architecture figure, walked in prose. Build the figure first and write the section to it; walk it left to right; give it a caption that re-tells the whole pipeline, self-contained, a reader of the caption alone should get the method.
  4. Star component fattest. The one genuinely new piece, the thing rivals lack, gets the most space and the most defense; the plumbing every rival also has is compressed.
  5. Running example threaded. Pick one concrete instance and lead every subsection with it before the general case, so intuition precedes each equation.
  6. Algorithm box with defaults. If there is a real procedure or loop, box it: titled with the system name, line-numbered, SMALL-CAPS subroutines, hyperparameters-with-defaults declared inside the box, each line cross-referenced to its subsection. Point at it; never transcribe it.
  7. Seal the section. No benchmark numbers, no results plot, no standalone related work, no limitations paragraph, those live in other sections. Cite prior work only instrumentally (the tool you reuse), never to survey a field.
  8. Close with a handoff. One forward-looking capability sentence, what the system can now do, not how well it scores, then straight into Experiments, pre-naming the ablation axis the reader will meet next.

Core rules

  • The star component (the one thing rivals lack) is the fattest, most-defended subsection; shared plumbing is compressed to make room for it.
  • Coin every component, bold it once, then use it as a proper noun, never fall back to "our iterative procedure."
  • Notation is a cost: introduce a tiny symbol set, define each at first use, state the object as a numbered display, and reuse it by number; never open a sentence with a symbol.
  • Credit the nearest rival warmly, then contrast on exactly one axis, never bash. Your closest rival is your likeliest reviewer.
  • Narrate the loop on one axis, per-module or per-timestep, and do not zig-zag; describe no mechanic twice on two axes.
  • Defend each design choice in one in-line clause, never in a rationale subsection.
  • Prose axioms: sentences <= ~28 words, subject and verb early, zero "can"/"could", no bare This/These, numbers paired and attached to exact objects.

Reference file

  • method-guide.md: full guide: term translations, the N-named-parts skeleton, three method recipes with fill-in templates, the devices recipe (figure and caption, algorithm box, running example, notation hygiene), mechanical checklists including the "never in a method section" list, the assembly order, and a worked example on a toy system.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.