SkillFoxx

PerforatedAI skill

Agent skills for Claude Code, Codex and Cursor that add artificial dendrites to a PyTorch model and analyze training results.

  • 241 ★ on the repository
  • Last commit
  • Apache-2.0
SkillOfficial

Extended access

Why this level

  • The skill edits the project's training script and run configuration.
  • Training stays with the user: the agent does not run code but reads the output it is sent.
All reasons and checks

perforatedai/perforatedaiAuthor:PerforatedAI

Install

Manual install

npx skills add PerforatedAI/PerforatedAI -a claude-code -s '*'

README command installing every skill into the project. Swap `claude-code` for `github-copilot`, `codex` or `cursor`.

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • The skill edits the project's training script and run configuration.
  • Training stays with the user: the agent does not run code but reads the output it is sent.

This is third-party code. Review the repository files before installing.

What it does

The `skills/` folder holds eight skills for wiring the PerforatedAI library into a PyTorch training script. The library adds artificial dendrites, extra learning signals at the level of individual neurons. The main `perforatedai` skill starts on the phrase "Perforate my model", reads the code and makes about ten lines of integration changes. The other skills load as needed. `perforatedai-analyze` reads training CSV output and recommends config changes, `perforatedai-distributed` fires on DataParallel and DDP, and `perforatedai-libraries-transformers` on the HuggingFace `Trainer`. `perforatedai-wandb` sets up Weights and Biases logging, `perforatedai-plot` draws accuracy against parameter count. The user starts training; the agent reads the output it is sent.

Who it is for. ML engineers and researchers on PyTorch who want to try dendritic optimization without hand-rewriting training code.

Good fit when

  • A PyTorch model has stalled on accuracy and you want to test dendrites on it.
  • Model size is capped and accuracy has to hold.
  • Training runs through the HuggingFace `Trainer` or on several GPUs.
  • After a run you need to read the CSV output and decide what to change in the config.

Not a fit when

  • The model is not written in PyTorch.
  • You only need library reference, with no script changes.
  • Training is tuned to a hard budget and the loop cannot change.

Example request

Perforate my model: wire dendrites into the training in `train.py` and tell me which command to run.

Limitations

Python 3.7 or newer and PyTorch 1.9 or newer are required. The skills do not install the package: run `pip install perforatedai perforatedbp` separately. Figures in the README refer to the full Perforated Backpropagation system in the Perforated AI Suite, available on request. The `npx skills add` command copies the skills into the chosen agent's folder. The skills are not meant for models built on other frameworks. The repository is licensed Apache-2.0.

How to disable. Delete the `perforatedai*` skill folders from `.claude/skills/` or the other agent's directory. Revert script edits with git.

FAQ

How do I install the PerforatedAI skills in Claude Code?

Open a terminal in the project root and run `npx skills add PerforatedAI/PerforatedAI -a claude-code -s '*'`. For another agent, swap the value after `-a` for `github-copilot`, `codex` or `cursor`.

What phrase starts the workflow?

"Perforate my model". Use "Debug my perforated model" to troubleshoot and "Export my perforated model" for ONNX, TFLite or TorchScript output.

What does the perforatedai-analyze skill do?

It reads the CSV files after training and recommends config changes. Invoke it with "Analyze my perforated results" once a run finishes.

Do I install the library separately?

Yes. The skills only edit the script, so run `pip install perforatedai perforatedbp` in the project environment before the first integration.

Do the skills cover multi-GPU training?

Yes. The `perforatedai-distributed` skill loads automatically when the main skill detects DataParallel or DDP.

Discussion

No questions or reviews yet.

Skills

Ledidi designs minimal edits to DNA, RNA and protein sequences for a target model output; its skill guides agents.

SkillMinimal access119

A template for data scientists: AGENTS.md, CLAUDE.md and skills for Python, SQL, Polars and model diagnostics.

SkillMinimal access115
Official

Probabl skills for machine learning experiments: a scikit-learn and skrub pipeline, evaluation, audit and a link to Skore Hub

PluginMinimal access145

17 skills for computational biology: RNA-seq, single-cell, protein structure, docking, literature and figures

SkillExtended access75

Around 200 skills for Claude Code on bioinformatics, chemistry and biostatistics, from RNA-seq analysis to molecular docking

SkillExtended access389

Скилл Claude Code учит агента библиотеке tangermeme для анализа геномных нейросетей: DeepLIFT/SHAP, мутагенез, дизайн.

SkillMinimal access333

README in short

The library adds dendrites to neurons of a PyTorch model and claims gains in accuracy, compression and data efficiency. The repository carries examples (MNIST, TD3, ImageNet ResNet-18, Edge Impulse) and API docs. The skills are covered in the AI Agent Skills section.

README badge

Original SKILL.md

---
name: perforatedai
description: "Expert in PerforatedAI library for adding artificial dendrites to PyTorch neural networks. Triggers: 'Perforate my model' (start interactive setup), 'debug my perforated model' (debug/optimize existing integration), 'load my perforated model for inference' (deploy trained models), 'export my perforated model' (ONNX/TFLite/TorchScript export). Also use when: debugging dendrite training, configuring PAI settings, troubleshooting dendrite issues, working with PAINeuronModule or PAIDendriteModule. For analyzing completed training results, use the perforatedai-analyze skill."
---

# PerforatedAI Dendrite Network Skill

## Available Resources

This skill references the following resources in the PerforatedAI repository on GitHub:
- **API Documentation**: [api/](https://github.com/PerforatedAI/PerforatedAI/tree/main/api) - All API guides and references
- **Source Code**: [perforatedai/](https://github.com/PerforatedAI/PerforatedAI/tree/main/perforatedai) - Complete PerforatedAI library implementation
- **Examples**: [examples/](https://github.com/PerforatedAI/PerforatedAI/tree/main/examples) - Working code examples for various architectures

Feel free to reference these when helping users debug or understand implementation details.

## Entry Points

### 1. "Perforate my model" - Interactive Setup

Report a problem