Machine learning

Generate, build, create, or design ComfyUI workflow JSON from natural language descriptions.

Use when the user requests diagrams, flowcharts, architecture diagrams, ER diagrams, UML / sequence / class diagrams, network topology, ML/DL model figures…

Create distinctive, production-grade frontend interfaces and printable HTML decks with high design quality.

Use when editing or reviewing existing prose for LLM writing habits, rhetorical padding, or fidelity to an author's voice.

Verification framework from the Stanford + NVIDIA LLM-as-a-Verifier paper (arXiv:2607.05391): continuous scores from expectation over scoring-token logits…

Local LLM inference setup and optimization for agentic workflows.

Run runtime-defined semantic decisions ("semantic ifs") locally with the SemIf CLI (`semif-score`): give a local LLM unstructured state + a criterion question…

Golang CLI application development.

Seed LLM GTM and competitive positioning.

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward…

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward…

RAG (Retrieval-Augmented Generation) system for affidavit workflows.

AI autonomous research agent for LLM training optimization using opencode as the agent.

Deterministic local code hygiene runner.

通用知识库构建 & 查询工具。给定一个包含 PDF/Sphinx-HTML 的原始文档文件夹，批量转成可 grep 的 Markdown、抽符号倒排、生成 LLM 摘要 sidecar 和 manifest 聚合索引。Agent 进入陌生领域时先 `ls ~/.maxclaw/kb/` 看有无相关知识库，再按"L3…

Seederive 非结构化数据打标平台，使用 LLM 对文本、语音、图片数据进行批量分析处理。 当用户提到以下任何场景时必须使用此 Skill：数据打标、标注、情感分析、标签分类、 观点提取、翻译、评论分析、水军识别、内容评分、标签库管理、提示词优化。…

Measures and improves the quality of AI models and agents on Google Cloud using the Eval Quality Flywheel methodology.

Clarify vague or underspecified creative intent before prompt construction.

Route prediction and forecasting problems to the right method.

Autonomous ML research loop for a "research lab" repo — survey, hypothesize, pre-register, run matched-compute experiments with seeds and a negative control…

Use when Codex needs to build or query a local RAG knowledge base for academic papers, literature matrices, public author guidelines, GB/T7714 rules, CSL…

Performs deep AI-powered analysis and understanding of long-form videos (up to 3 hours, 10GB) using Volcengine LAS large language models.

Build an automated evaluation pipeline that tests a Python LLM application end-to-end with pixie test — real code paths, real LLM calls, instrumented external…

检索多模态数据处理平台已完成工作流的最新处理结果。适用于网页爬取、文档解析、OCR、Markdown/JSON 转换等已写入数据集的结果问答，并返回可回溯的数据集版本与来源引用。