Kang Heng
Community@Nzettodess
Kang Heng's registry of 137 skills spanning LLM inference serving, fine-tuning, quantization, RAG, evaluation, data engineering, and developer productivity utilities.
Agent Skills by Kang Heng
Showing 137 vetted skills indexed across 1 GitHub repositories.
Nzettodessskill-creator
Create, evaluate, and iteratively improve AI agent skills with benchmarked testing workflows.
Nzettodessstorage-analyzer
Scans macOS and Windows disk usage and generates an interactive HTML cleanup report.
Nzettodessaihot
Fetches Chinese AI news briefings from the aihot.virxact.com public REST API.
Nzettodesscanvas-design
Create original visual art and posters as PNG or PDF files using design philosophies.
Nzettodessfrontend-ui-ux
Generates production-grade frontend interfaces with distinctive typography, color, and motion design.
Nzettodessponytail
Enforces minimal, standard-library-first solutions for coding tasks with adjustable intensity levels.
Nzettodessponytail-audit
Audits an entire codebase for over-engineering and ranks what to delete or simplify.
Nzettodessponytail-help
Displays a quick-reference card for ponytail modes, skills, and commands.
Nzettodessponytail-review
Reviews code diffs for over-engineering and lists what to delete or simplify.
Nzettodessponytail-gain
Displays a one-shot ASCII scoreboard of ponytail's benchmarked code, cost, and speed reductions.
Nzettodessponytail-debt
Collects ponytail comment markers from a codebase into a technical debt ledger.
Nzettodessmermaidjs-v11
Create diagrams using Mermaid.js v11 syntax and render them to SVG, PNG, or PDF.
Nzettodessplotly
Create interactive scientific and statistical charts in Python with Plotly.
Nzettodessseaborn
Generate statistical visualizations in Python including scatter plots, heatmaps, violin plots, and faceted figures.
Nzettodessreadme
Generates comprehensive README.md documentation by exploring codebase structure, configuration, and deployment targets.
Nzettodessneat-freak
Reconciles project documentation and agent memory against code after development sessions.
Nzettodesssetup-matt-pocock-skills
Configures issue tracker, triage labels, and domain docs for engineering skills.
Nzettodessimprove-codebase-architecture
Identifies shallow modules and proposes deepening refactors rendered as an HTML architecture review report.
Nzettodesstriage
Triage GitHub issues through a state machine of category and state roles.
Nzettodessprototype
Build throwaway terminal or UI prototypes to validate logic and design questions.
Nzettodessdiagnose
Diagnose hard bugs and performance regressions through a structured reproduce-hypothesise-fix loop.
Nzettodessto-issues
Breaks plans and PRDs into vertical-slice issues on the project issue tracker.
Nzettodesszoom-out
Maps relevant modules and callers to explain unfamiliar code at a higher abstraction level.
Nzettodessgrill-with-docs
Challenges plans against domain glossaries and updates CONTEXT.md and ADR documentation inline.
Frequently Asked Questions About Kang Heng
FAQPage SchemaWhat tasks can I accomplish with Kang Heng's skill registry?βΌ
You can serve LLMs with vLLM, TensorRT-LLM, SGLang, or llama.cpp; fine-tune models via PEFT, Unsloth, Axolotl, and TRL; quantize with GPTQ, AWQ, HQQ, and bitsandbytes; build RAG systems with Chroma, Pinecone, Qdrant, and FAISS; plus run data analysis, Power BI modeling, and Mermaid/Plotly visualization.
Who is the target audience for these skills?βΌ
ML engineers deploying production inference, researchers fine-tuning or evaluating models, data scientists running statistical analysis with pandas, statsmodels, and scikit-survival, and developers seeking code-simplification reviews, TDD loops, issue triage, and documentation cleanup utilities.
How do I install and run these skills?βΌ
Each skill ships as a folder with native frontmatter declaring name, description, and dependencies. Python-based skills list pip-installable packages such as vllm, torch, transformers, peft, or chromadb; invoke them through a compatible agent host like Claude Code, OpenAI Codex, OpenCode, or OpenClaw.
Are these skills open source and what do they cost?βΌ
Most skills carry the MIT license, including the Orchestra Research ML collection, ponytail suite, and strategy-advisor. Exceptions exist: xlsx and canvas-design use proprietary terms detailed in their LICENSE.txt files. The skills themselves are free; underlying GPU cloud or model API usage may incur provider costs.
What prerequisites and dependencies are required?βΌ
ML skills require PyTorch and HuggingFace Transformers, with specific extras like flash-attn, deepspeed, ray, or tensorrt-llm per skill. GPU skills assume NVIDIA hardware (A100/H100) while llama.cpp runs on CPU and Apple Silicon. Power BI skills need the power-bi-modeling MCP connection; aihot needs no API key.