AI & LLM Integration for Claude Code

Guidelines for testing the application with Vitest, including unit tests, integration tests (emulator), AI tests, and eval suites for LLM features

Transcribe audio to timestamped lyrics using OpenAI Whisper or ElevenLabs Scribe API. Outputs LRC, SRT, or JSON with word-level timestamps. Use when users want to transcribe songs, generate LRC files, or extract lyrics with timestamps from audio.

Build agents that speak your UI

Compiles any research input — PDF papers, GitHub repositories, experiment logs, code directories, or raw notes — into a complete Agent-Native Research Artifact (ARA) with cognitive layer (claims, concepts, heuristics), physical layer (configs, code stubs), exploration graph, and grounded evidence. Use when ingesting a paper or codebase into a structured, machine-executable knowledge package, building an ARA from scratch, or converting research outputs into a falsifiable, agent-traversable form.

Evaluates NVIDIA Cosmos Policy on LIBERO and RoboCasa simulation environments. Use when setting up cosmos-policy for robot manipulation evaluation, running headless GPU evaluations with EGL rendering, or profiling inference latency on cluster or local GPU machines.

Provides guidance for automatically evolving and optimizing AI agents across any domain using LLM-driven evolution algorithms. Use when building self-improving agents, optimizing agent prompts and skills against benchmarks, or implementing automated agent evaluation loops.

Provides guidance for experiment tracking with SwanLab. Use when you need open-source run tracking, local or self-hosted dashboards, and lightweight media logging for ML workflows.

Fine-tunes and evaluates OpenVLA-OFT and OpenVLA-OFT+ policies for robot action generation with continuous action heads, LoRA adaptation, and FiLM conditioning on LIBERO simulation and ALOHA real-world setups. Use when reproducing OpenVLA-OFT paper results, training custom VLA action heads (L1 or diffusion), deploying server-client inference for ALOHA, or debugging normalization, LoRA merge, and cross-GPU issues.

Fine-tune and serve Physical Intelligence OpenPI models (pi0, pi0-fast, pi0.5) using JAX or PyTorch backends for robot policy inference across ALOHA, DROID, and LIBERO environments. Use when adapting pi0 models to custom datasets, converting JAX checkpoints to PyTorch, running policy inference servers, or debugging norm stats and GPU memory issues.

Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning neural networks, debugging loss spikes or OOM, choosing architectures, or optimizing GPU throughput.

LLM architectures and implementations including LitGPT, Mamba, NanoGPT, RWKV, and TorchTitan. Use when implementing, training, or understanding transformer and alternative architectures.

Meta's 86M prompt injection and jailbreak detector. Filters malicious prompts and third-party data for LLM apps. 99%+ TPR, <1% FPR. Fast (<2ms GPU). Multilingual (8 languages). Deploy with HuggingFace or batch processing for RAG security.

Use when fine-tuning LLMs, training custom models, or adapting foundation models for specific tasks. Invoke for configuring LoRA/QLoRA adapters, preparing JSONL training datasets, setting hyperparameters for fine-tuning runs, adapter training, transfer learning, finetuning with Hugging Face PEFT, OpenAI fine-tuning, instruction tuning, RLHF, DPO, or quantizing and deploying fine-tuned models. Trigger terms include: LoRA, QLoRA, PEFT, finetuning, fine-tuning, adapter tuning, LLM training, model t

Designs and implements production-grade ML pipeline infrastructure: configures experiment tracking with MLflow or Weights & Biases, creates Kubeflow or Airflow DAGs for training orchestration, builds feature store schemas with Feast, deploys model registries, and automates retraining and validation workflows. Use when building ML pipelines, orchestrating training workflows, automating model lifecycle, implementing feature stores, managing experiment tracking systems, setting up DVC for data vers

> EU AI Act per-system inventory — track each AI system's role (provider, deployer, importer, distributor, authorized representative, product manufacturer) and risk tier (prohibited, high-risk, limited, minimal, GPAI, GPAI+systemic). Role and tier are assessed per system, not per company. Use when the user says "ai inventory", "add an ai system", "what systems do we have", "classify this ai system", "eu ai act register", or "ai system registry".

> Detects when Luminance, Kira, or a similar bulk-review tool is in use, hands off the high-volume clause extraction to it, and QAs its output per the trust level in `~/.claude/plugins/config/claude-for-legal/corporate-legal/CLAUDE.md`. Use when user says "send to Luminance", "bulk review", "AI extraction", or when diligence-issue-extraction hits a high-volume category.

> Run an AI impact assessment — structured intake, risk analysis, regulatory classification per regime in scope, policy consistency diff, and recommendation with conditions. Uses the house-style structure learned from the seed impact assessment in `~/.claude/plugins/config/claude-for-legal/ai-governance-legal/CLAUDE.md`. Use when user says "impact assessment for", "assess this AI use case", "run an AIA", "generate an AIA", "we need to document this AI system", "AI risk assessment for X", or foll

> Draft a firm AI usage policy from published model policies, adapted to your practice profile — a research-and-synthesis tool whose output is a draft for attorney review and adoption, not a finished policy. Use when user says "draft an AI policy", "we need an AI policy", "build an AI usage policy", "our firm needs a GenAI policy", or similar requests to generate a first-cut internal AI policy.

> Classify a proposed AI use case against your registry — approved, conditional, or not approved — and produce required conditions and next steps. Flags cross-plugin handoffs to privacy or product counsel. Use when user says "triage this use case", "can we use AI for X", "is this approved", "what do we need to do to use AI for X".

> Review vendor AI terms — agreement, addendum, or ToS AI provisions — against your governance positions; flag training-on-data, liability, model changes, and AI policy consistency. Use when user says "review this AI agreement", "check OpenAI terms", "what did we agree to with [vendor]", "vendor sent an AI addendum", "is this AI contract okay", or attaches vendor AI terms.

| digoal/德哥的思维框架与表达方式。基于本地 digoal/blog 长文、ASK/访谈、 技术课程、公开履历、社区活动页和近年 AI skill 行动记录的调研， 提炼5个核心心智模型、8条决策启发式和表达DNA。 Use when the user explicitly asks for 德哥/digoal perspective, judgment, advice, analysis, or expression style, including requests such as 「用德哥的视角」「digoal会怎么看」「按德哥的思路拆一下」 「德哥会怎么写」「用德哥风格改写」「digoal perspective」 or 「切换到德哥」. Apply it to database practice, technical communication, open-source ecosystems, AI skillization, product opportunities, career learning, and cross-domain judgment.

每次启动一位领域专家时,把下面模板里的占位符替换掉,作为 prompt 喂给 Agent 工具。

召唤多领域专家智能体，对任何棘手、跨学科、有争议或开放性的问题做并行深度分析，再经过一轮克制的事实核查与红队反驳，最后由一个"干净大脑"的终稿撰写者把所有内容无缝重写成一篇第一人称、面向小白、图文并茂、有人味、留有证伪空间的深度长文。触发条件：用户提出任何值得深挖的复杂问题，尤其是跨领域、需要多方视角、需要权威数据支撑的问题；提到"帮我深度分析"、"多专家视角"、"专家怎么看"、"这个问题涉及哪些领域"、"帮我搞清楚 XX 到底是怎么回事"、"用第一性原理分析一下"、"这个观点站得住脚吗"、"帮我论证一下"、"multi-expert-analyzer2"，或者只是抛出一个开放性问题如"为什么 XX 会发生"、"如何看待 XX 现象"、"XX 到底靠不靠谱"。即使用户没有明确说"专家"或"深度分析"这几个字，只要问题本身跨越多个专业领域、需要审慎论证而非泛泛而谈，也应使用本 skill。本 skill 与仓库中其他类似名称的技能（如 multi-expert-analyzer、multi-expert-analyzer1）相互独立，不共享模板或输出结构，请完全按照本文件的流程执行。

Answer general or cross-domain questions with a non-pleasing rational mode: adversarial red-team and blue-team expert analysis, mutually exclusive conclusions, up to five debate rounds, saved Markdown intermediate expert outputs, and a final first-person judge-written plain-language article. Use when the user asks for rational adversarial analysis, red/blue team debate, non-flattering judgment, evidence-backed multi-expert reasoning, or a final Markdown answer to a broad question.

所有专家稿落盘后再启动红队,让它找到最强主张并尝试反驳。

Runs a budget-aware agent-assisted engineering health audit for instruction/config drift, hooks/MCP, verifier surfaces, and AI maintainability. Use when users ask in any language to audit Claude, Codex, Pi, agent instructions, MCP or hooks, verifier coverage, or AI-maintainability drift. Not for debugging application code or reviewing PRs.

번개장터 검색, 상세조회, 찜, 채팅, 대량 수집, AI TOON export를 bunjang-cli로 안내한다.

AI가 쓴 티가 나는 한국어 글을 자연스러운 사람 글로 고친다. 번역체, AI 상투어, 과도한 명사화·피동, 3의 법칙, 과장된 의의 부여, 마무리 상투구, 챗봇 잔재, 줄표·곡선따옴표 같은 한국어 특유의 AI 흔적을 심각도(S1/S2/S3)로 분류해 잡아내고 의미는 보존하면서 다시 쓴다. 목표 글자수를 함께 주면(예: "1000자로", length=1000) 그 분량에 맞춰 늘리거나 줄인다. "AI 티 안 나게", "사람이 쓴 것처럼", "자연스럽게 다듬어줘", "번역체 고쳐줘", "어색한 거 고쳐줘", "N자로 맞춰서" 같은 요청에 사용.

Deduplication judge for the c-review pipeline. Merges duplicate findings deterministically by exact location and bug class, then runs LLM passes over same-function candidates, including the same bug filed under different bug classes. Spawned by the c-review skill orchestrator only.

Stops, cancels, or aborts the active skill improvement loop while preserving all changes made to skill files. Use to manually stop, cancel, abort, or kill the iteration process early.

Deduplication judge for the rust-review pipeline. Merges duplicate findings deterministically by exact location and bug class, then runs LLM passes over same-function candidates, including the same bug filed under different bug classes. Spawned by the rust-review skill orchestrator only.

Runs external LLM code reviews (OpenAI Codex or Google Gemini CLI) on uncommitted changes, branch diffs, or specific commits. Use when the user asks for a second opinion, external review, codex review, gemini review, or mentions /second-opinion.

Distill an ex-partner into an AI Skill. Import WeChat history, photos, social media posts, generate Relationship Memory + Persona, with continuous evolution. | 把前任蒸馏成 AI Skill，导入微信聊天记录、照片、朋友圈，生成 Relationship Memory + Persona，支持持续进化。

Assess whether your product work is AI-first or AI-shaped. Use when evaluating AI maturity and choosing the next team capability to build.

Diagnose context stuffing vs. context engineering. Use when an AI workflow feels bloated, brittle, or hard to steer reliably.

Identify what to measure, test, or track to de-risk a product or AI idea. Use when stress-testing an idea across internal (DUFV) and external (PESTEL) dimensions.

Evaluate an AI product idea across outcomes, hypotheses, risks, and positioning. Use when deciding whether an AI solution deserves investment or recommendation.

Generate AI videos using Volcengine Seedance model. Supports text-to-video (T2V), image-to-video (I2V), and audio-synced video generation. Use this skill when the user wants to create or generate videos.

Generate AI images using Volcengine Seedream model. Supports text-to-image (T2I), image editing (I2I), multi-image fusion, and web-search-based generation. Use this skill when the user wants to create, generate, or edit images.