agentleFS
Sign inStart free
hasantahabagcihasantahabagci/agentic-paper2code/.claude/skills/paper2code/SKILL.md

paper2code

Skill0 starsChanged 2 months agoApache-2.0View on GitHub

Reproduce a research paper as a working code repository. Use whenever the user wants a paper implemented, reproduced, or turned into code - "implement this arXiv paper", "reproduce this method", "build the model from this PDF", "port this paper to PyTorch". Runs a ten-stage pipeline (ingest, plan, design, tasks, config, analyze, implement, verify, review, package) that produces annotated artifacts, a repository that actually runs, and an honest record of what was and was not verified.

5 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions5 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 5

    1. PrathamLearnsToCode/paper2code★ 1.5ksame name
    2. lyx2022518/sherlock-ai-plugin★ 3same name
    3. dldush/claude-skill-set★ 2same name
    4. SCUTxyx/paper2code★ 1same name
    5. aresbit/claude-skills★ 0same name

    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 repository3 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 1

    Distinctive

    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. Very specific:

    Read prompts/shared/rules.md before anything else.

    line 18
  2. Very specific:

    Never invent a number: Every hyperparameter is annotated # paper §4.1, # external: <source>, or # UNSPECIFIED - <why this value>.

    line 22
  3. Very specific:

    State lives on disk: Update .p2c/<slug>/STATE.md after every step, so the work survives a lost context window.

    line 27
  4. Very specific:

    Read each prompt in full and execute it in order: 00-ingest, 01-plan, 02-design, 03-tasks, 04-config, 05-analyze, 06-implement, 07-verify, …

    line 34
  5. Distinctive:

    After stage 1 (are these the right reproduction targets?), after stage 4 (how much did the paper leave unspecified?), and after stage 7 …

    line 51

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “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

Reproducing a paper as code

This skill runs the agentic-paper2code pipeline. It exists because the usual approach — handing a PDF to a model and asking for an implementation — produces code that looks right, has never run, and silently invents the hyperparameters the paper did not state.

First, read the contract

Resolve the toolkit root: the first of ${CLAUDE_PLUGIN_ROOT}, .p2c/toolkit, the repository root, or ~/.paper2code that contains a prompts/ directory.

Read prompts/shared/rules.md before anything else. Four rules matter most:

  1. Fidelity beats cleverness. Implement the paper as written. Objections go in notes.md; the code follows the paper.
  2. Never invent a number. Every hyperparameter is annotated # paper §4.1, # external: <source>, or # UNSPECIFIED - <why this value>. An unannotated value is a defect.
  3. No stubs. No TODO, no placeholder bodies, no training loop that never steps the optimizer.
  4. State lives on disk. Update .p2c/<slug>/STATE.md after every step, so the work survives a lost context window.

Then read prompts/shared/workspace.md for the directory layout.

Then run the stages

Read each prompt in full and execute it in order: 00-ingest, 01-plan, 02-design, 03-tasks, 04-config, 05-analyze, 06-implement, 07-verify, 08-review, 09-package, all under prompts/.

Two properties of the pipeline are worth protecting:

  • Stage 5 is separate from stage 6 on purpose. Reasoning about masking logic and managing imports at the same time is how subtle errors get in. Analyse every file completely before writing any code.
  • Stage 7 is what makes this different from one-shot generation. You can install dependencies, run the smoke config, overfit a single batch, and fix what breaks. Code that has never executed is a hypothesis; use the tools.

Parallelise stage 5 with subagents, one per file. Never parallelise stage 6.

Check in with the user

After stage 1 (are these the right reproduction targets?), after stage 4 (how much did the paper leave unspecified?), and after stage 7 (does it run, does it learn?). Between those, keep going.

Report honestly

The deliverable includes what failed. Report the fidelity score, the open questions, the reproduction targets you could not run, and the numbers you got instead — including the disappointing ones. A reproduction that reports 24.1 against the paper's 27.3 is a useful result. One that claims 27.3 without having run is worthless.

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.

Check your setup, free