Skip to content

AI & Agentic Coding Skills

This repository includes a suite of specialized AI Agent Skills located in agents/skills/. These skills provide structured workflows for AI coding assistants (e.g., Antigravity, Claude, Copilot) to automate complex Deep Learning engineering tasks.

Available AI Skills Overview

Skill Name Purpose & Function Location
convert-to-dl-template Converts any legacy Python/DL codebase to strictly match this boilerplate architecture agents/skills/convert-to-dl-template/
dl-experiment-runner Automates hyperparameter configuration, training execution, and metric reporting agents/skills/dl-experiment-runner/
dl-model-exporter Exports trained PyTorch checkpoints (.pt) into ONNX & TorchScript production formats agents/skills/dl-model-exporter/
dl-paper-to-code Translates novel academic paper architectures and equations into modular PyTorch components agents/skills/dl-paper-to-code/

🛠️ Skill Details & Usage

1. convert-to-dl-template

  • Trigger: When converting a legacy repository or setting up a new project from raw scripts.
  • Actions:
  • Reorganizes raw PyTorch models into src/models/ and sub-blocks into src/backend/components/.
  • Refactors all imports to absolute src. paths.
  • Generates build configuration (pyproject.toml, mkdocs.yml, CITATION.cff) and CI/CD pipelines.

2. dl-experiment-runner

  • Trigger: When running hyperparameter sweeps, benchmarking, or comparing model variants.
  • Actions:
  • Generates YAML config overrides in inputs/experiments/.
  • Executes python scripts/train.py --config ....
  • Parses logs and outputs comparative Markdown tables with loss and accuracy progressions.

3. dl-model-exporter

  • Trigger: When preparing trained models for deployment.
  • Actions:
  • Loads PyTorch weights from outputs/weights/best_model.pt.
  • Traces forward pass with dummy inputs and exports to ONNX/TorchScript in outputs/artifacts/.
  • Verifies ONNX runtime inference correctness.

4. dl-paper-to-code

  • Trigger: When translating a paper, equation, or pseudo-code into PyTorch.
  • Actions:
  • Breaks down paper modules into layers (src/backend/components/layers/), blocks (blocks/), and stages (stages/).
  • Assembles the complete model wrapper in src/models/.
  • Writes unit tests in tests/unit/ to verify tensor shapes and parameter counts.