agentleFS
mkaremm80mkaremm80/-my-claude-skills/skills/paper-from-zero-guide/SKILL.md

paper-from-zero-guide

Skill1 starsChanged 12 days agoMITView on GitHub

Turn any research paper (PDF or LaTeX) into a colorful PDF guide that explains it from zero: background, every equation with labels, full proofs, examples, results, and simulation checks.

  • Installs packages

1 other project uses a version of this, 1 of them edited.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 1

    1. mkaremm80/My-Skills-on-Claude★ 5289% same

    Wrote their own version · 0

    None found.

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository5 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 4 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use a task list: (1) map the paper, (2) compute examples + verify, (3) helper drawings, (4) write + compile, (5) visual check + deliver.

    line 12
  2. Very specific:

    If a zip: unzip into its own folder.

    line 15
  3. Very specific:

    Never name a script numbers.py (shadows a stdlib module).

    line 22
  4. Distinctive:

    Use the paper's nominal operating point and one fixed-seed realization for examples.

    line 23
  5. Distinctive:

    Verify every closed-form result: Monte Carlo (~1e6 samples) or direct computation.

    line 25
  6. Distinctive:

    Do the proofs yourself on paper first and check them numerically (e.g., bound >= true value for the paper's own configuration).

    line 27
  7. Very specific:

    Put \Needspace{14\baselineskip} before every \section (a regex pass is easiest) and \clearpage before each Part and before the summary, to …

    line 80
  8. Very specific:

    Avoid long lists of numbers inside $...$ in narrow table cells (they cannot break and overflow).

    line 81

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Paper From Zero Guide

Make one colorful A4 PDF (about 20-30 pages) that lets a reader with ZERO background understand a research paper fully: the field basics, every equation, every proof, every figure/table, and whether the paper's numbers are right.

Writing style: English, short sentences, simple words (B2 level), professional. Numbers quoted from the paper must match the paper exactly. Numbers from your own worked examples must be labeled "Example" with their parameters and seed. Deliver as <Short_Paper_Name>_Guide_from_Zero_v1.pdf; number later versions v2, v3, ...

Use a task list: (1) map the paper, (2) compute examples + verify, (3) helper drawings, (4) write + compile, (5) visual check + deliver.

Step 1 - Read and map the paper

  1. If a zip: unzip into its own folder; read main.tex, all \input files (tables, TikZ figures), references.bib, figure scripts and data files (e.g. figdata.json, plots.py). If a PDF: extract text and look at each figure page.
  2. List: every numbered equation, every proposition/lemma/proof, key unnumbered formulas (scores, rules, complexity, metrics such as % reduction, confidence intervals, gap closed), every symbol with meaning and nominal value, all reported numbers in text/tables/figures.
  3. Build the equation map: which equation feeds which (model -> estimator -> score -> decision; geometry -> dimension -> rule; theory -> the main rule -> results).
  4. List the background concepts a beginner needs (e.g., for wireless: multipath fading, complex gain, SNR/dB, Gaussian/complex Gaussian, covariance/correlation, eigenvalues, rank/nullspace, singular values, pseudoinverse, projector, Bayesian prior/posterior, outage). Pick only what this paper uses.
  5. Note notation problems and small errors for the authors: one symbol used for two things, undefined symbols, missing subscripts, conditions a proof silently needs (e.g., full column rank), examples in the text that do not match the stated rule, tie-breaking or other rules left unstated.

Step 2 - Compute worked examples and verify (Python, numpy/scipy)

  • Never name a script numbers.py (shadows a stdlib module). Use eq_numbers.py, save results to numbers.json.
  • Use the paper's nominal operating point and one fixed-seed realization for examples. For each equation compute concrete values (matrix entries, posterior mean/variance at a few indices, scores, probabilities, eigenvalues and cumulative energy, singular values, bound values, operation counts).
  • Find a teaching example that shows WHY an equation matters (e.g., same mean but different variance gives opposite outage behavior; greedy choice clustering).
  • Verify every closed-form result: Monte Carlo (~1e6 samples) or direct computation. Recompute every number the paper reports that can be recomputed (eigenvalues, dimensions, condition numbers, errors) and record paper vs. ours.
  • If the paper has a main simulation, write an independent small re-simulation of the key curve(s) (e.g., 20,000 trials, common random numbers). Run long jobs with nohup python3 script.py args > log 2>&1 & in the background and keep working; save results to JSON. Compare with the paper's figure data and plot ours (lines) vs paper (circles).
  • Do the proofs yourself on paper first and check them numerically (e.g., bound >= true value for the paper's own configuration).

Step 3 - Helper drawings (matplotlib)

Settings: font.family STIXGeneral, mathtext.fontset stix, pdf.fonttype 42, top/right spines off, light grid, about 6.6 x 3.0 in, 10-11 pt. Save each as PDF (for LaTeX) and PNG at 70 dpi (for checking). Notes: ax.annotate(text, xy, xytext, arrowprops=dict(arrowstyle='-|>', color='#C0392B'), bbox=dict(boxstyle='round,pad=0.3', fc='#FFF8E1', ec='#E09F3E')). Put notes in empty areas, raise ylim for headroom, keep legends off the data. Typical set (adapt to the paper):

  • what the raw phenomenon looks like for one realization (e.g., power across positions with a threshold),
  • the model equation's meaning (e.g., correlation vs distance for several parameters),
  • random variables as clouds in a plane with the failure region shaded (shows a probability formula visually),
  • what the estimator believes after 1 measurement and after B measurements, for each policy (true vs estimate +- band, probed points marked, chosen item circled),
  • the greedy/adaptive policy step by step (3 small panels),
  • a probability formula vs its inputs with the teaching example marked,
  • spectrum + cumulative energy (dimension definition), and mode shapes,
  • our re-simulation vs the paper's values. Make a vertical montage of all PNGs and LOOK at it; fix overlaps and wrong statements in notes before continuing.

Step 4 - Labeled equations (annotate-equations)

\usepackage{annotate-equations}
\renewcommand{\eqnhighlightheight}{\mathstrut}
\renewcommand{\eqnannotationfont}{\bfseries\small}
\begin{eqbox}{Equation (3) -- best guess after t probes}
\vspace*{2.6em}
\begin{equation*}
\eqnmarkbox[teal]{q3mu}{\bm\mu}=\eqnmarkbox[navy]{q3G}{\mathbf G}\,\eqnmarkbox[amber]{q3A}{\mathbf A^{-1}}\,\eqnmarkbox[rosered]{q3y}{\mathbf y}
\end{equation*}
\annotate[yshift=1.3em]{above,left}{q3mu}{best guess}
\annotate[yshift=-1.3em]{below,right}{q3y}{measured data}
\vspace*{2.4em}
\end{eqbox}

Rules (learned the hard way):

  • Mark names must be UNIQUE across the whole document: prefix them per box (q1..., q2...). Easiest: write plain names, then run a Python pass that prefixes names inside each eqbox block with q<k>.
  • Alternate above/below and left/right. Two labels on the same side at the same height WILL collide: stack them (yshift 1.3em, 2.7em, 3.9em) and enlarge the \vspace* before/after accordingly (about 4-5em when stacked).
  • Two-line boxes (e.g., mean then covariance): leave about 5em between the lines, or the lower labels of line 1 hit the upper labels of line 2.
  • Labels near the box edge: use left labels for left terms and right labels for right terms, keep texts 2-5 words, shorten if they cross the frame.
  • A label line should not pass through another label's text; if it does, move one label to the other side or another height.
  • Needs at least 2-3 pdflatex runs.

Step 5 - Build the PDF (LaTeX article)

Preamble: [11pt,a4paper]article, geometry margin 2cm, helvet sans as default family, amsmath,amssymb,bm,mathtools, xcolor[table], tcolorbox[most], tikz with libraries arrows.meta,positioning,shapes.geometric,calc,fit,backgrounds,decorations.pathreplacing,decorations.pathmorphing (pathmorphing is needed for snake), booktabs,tabularx,xltabular,array,enumitem,titlesec,fancyhdr,float,caption,needspace,graphicx,annotate-equations,hyperref. Colors: navy #1B3A5C, teal #00798C, rosered #D1495B, amber #E09F3E, leaf #2E8B57, plum #6A4C93. Boxes (all enhanced,breakable, title={#1}): eqbox (white, navy frame: labeled equation), verdict (teal: key takeaways), goodbox (leaf: small example with numbers), notebox (amber: notes, cautions, notation points), proofbox (plum: full proofs, numbered steps, end with a black square), zerobox (rosered: everyday analogy). Column types: \newcolumntype{P}[1]{>{\raggedright\arraybackslash}p{#1}} and \newcolumntype{Y}{>{\raggedright\arraybackslash}X}; use Y instead of X so narrow columns do not stretch. Header: paper short title left, "Guide from zero -- v1" right; page number center. Section titles navy with a teal rule.

Structure:

  1. Cover: navy band via TikZ overlay (title "Understanding the Paper from Zero", paper title, authors, version, date), then a verdict box "The whole paper in five sentences", a notebox "How to read this guide" (say which numbers are examples), and a small TikZ roadmap of the parts. Then table of contents.
  2. Part A - Background from zero: one short section per needed concept, each with a picture or analogy (zerobox). Include a linear-algebra/probability tools table (tool | meaning in simple words | where used). End with the paper's problem as an analogy.
  3. Part B - Map and symbols: TikZ equation map on an absolute grid (\def\dx{40mm}\def\dy{25mm}, nodes about 29 mm wide, straight arrows, no crossings, short arrow labels), one-line story, "how each section is built" note; all-symbols table (symbol | simple meaning | value in paper); notation-points notebox.
  4. Part C - Every equation, with proofs: one section per equation/proposition: labeled eqbox -> In plain words -> proofbox with a full step-by-step proof (state every condition the proof needs) -> helper drawing -> goodbox small example with real numbers (and Monte Carlo check) -> Why the paper needs it. Also cover policy/score rules, the main proposed rule (re-use the paper's own TikZ system figure via \resizebox{0.97\textwidth}{!}{\input{...}} if it exists), and complexity with an operation count.
  5. Part D - Results explained: experiment setup list; metrics table with formulas (e.g., % reduction, CI = mean +- t*s/sqrt(n) with the t value, gap closed) and one paper example each; rebuild the main tables as colored tables (\rowcolors) with "how to read it"; include the paper's figures (\includegraphics of the PDFs) each followed by a notebox "What to see"; the mechanism explanation; limitations stated by the authors.
  6. Part E - Checks and summary: "Checked by our own simulation" table (item | formula/paper | ours | OK), our re-simulation figure, a verdict box on correctness listing any errors/notation points; summary table (one line per equation) using tabularx (NOT xltabular: an xltabular ending the document caused an \cmrsideswitch error); mini glossary. Put \Needspace{14\baselineskip} before every \section (a regex pass is easiest) and \clearpage before each Part and before the summary, to avoid orphan headings. Avoid long lists of numbers inside $...$ in narrow table cells (they cannot break and overflow); write them as text with $-$ for minus signs.

Compile 3 times: pdflatex -interaction=nonstopmode. Check the log for ^! errors and Overfull lines and fix them. Missing packages: apt-get install -y texlive-latex-extra texlive-fonts-extra texlive-science texlive-pictures.

Step 6 - Check before sending

  • Render all pages with pdftoppm -r 45 -png and make a montage (7 per row); look at page breaks, empty areas, orphan headings.
  • Render the equation pages at 75 dpi (4 pages side by side) and inspect EVERY eqbox: no overlapping labels, no label crossing the frame, no line crossing label text. Fix and recompile until clean. (pdftoppm is enough; installing PyMuPDF can be slow.)
  • Confirm all paper numbers match the paper and all example numbers match numbers.json.
  • Deliver the PDF with SendUserFile (display: render). In the reply, say in a few short lines what each part contains, what was verified (and how closely our re-simulation matches), and list the notation points/errors found for the authors.

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.