Lejzer

The registry covers LLM fine-tuning (LoRA, DPO, GRPO), distributed training (FSDP, DeepSpeed, torchtitan), inference serving (vLLM, TensorRT-LLM, llama.cpp), vector search (Qdrant, Pinecone, FAISS, Chroma), image/video/music generation, Excel financial models, GitHub operations, web pentesting, and macOS/desktop integrations.

ML engineers training and deploying language models, full-stack developers building with shadcn, Cloudflare Workers, or the Vercel AI SDK, security researchers running authorized pentests and OSINT, finance analysts building DCF/LBO models, and creatives producing video, comics, and generative art.

Skills are folder-based SKILL.md packages loaded by the Hermes Agent runtime on Linux, macOS, or Windows. Many declare pip dependencies (torch, transformers, vllm) or CLI prerequisites (docker, gh, ffmpeg, comfy-cli) that must be installed before invocation.

Nearly all skills carry the MIT license, with several finance and video skills under Apache-2.0. One exception is the powerpoint skill, which is proprietary with terms in its LICENSE.txt. External services like Pinecone, Lambda Labs, or Stripe may incur their own usage costs.

ML skills require PyTorch, Transformers, and often CUDA GPUs; finance skills need openpyxl; integration skills require API tokens (Shopify, Notion, Airtable, Google OAuth2, 1Password service accounts); macOS skills need CLIs like imsg, memo, or remindctl. Each frontmatter lists exact dependencies.