Top Agent Skills

Redige o agravo do CPC 1.042 contra a decisao que INADMITE REsp ou RE na origem (salvo inadmissao por repercussao geral/repetitivo, caso em que cabe agravo interno) e os embargos de divergencia do CPC 1.043-1.044 (acordao de orgao fracionario do STJ/STF divergente). Use quando o operador disser agravo em recurso especial/extraordinario, agravo do 1.042, REsp/RE inadmitido, nao subiu o recurso, embargos de divergencia, divergencia entre turmas do STJ/STF.

Faz o juizo de admissibilidade recursal completo antes de interpor: cabimento, tempestividade (dias uteis + dobro + feriado local da Lei 14.939/2024), preparo/desercao (CPC 1.007), regularidade formal, legitimidade/interesse e sucumbencia. Gate transversal de TODO recurso. Use quando o operador disser esse recurso e cabivel?, vai ser conhecido?, conferir admissibilidade, checar preparo, desercao, feriado local, interesse recursal, antes de protocolar o recurso.

Managed vector database for production AI applications. Fully managed, auto-scaling, with hybrid search (dense + sparse), metadata filtering, and namespaces. Low latency (<100ms p95). Use for production RAG, recommendation systems, or semantic search at scale. Best for serverless, managed infrastructure.

Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.

GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flexible quantization from 2-8 bit without GPU requirements.

Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.

LLM observability platform for tracing, evaluation, and monitoring. Use when debugging LLM applications, evaluating model outputs against datasets, monitoring production systems, or building systematic testing pipelines for AI applications.

Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform

Framework for state-of-the-art sentence, text, and image embeddings. Provides 5000+ pre-trained models for semantic similarity, clustering, and retrieval. Supports multilingual, domain-specific, and multimodal models. Use for generating embeddings for RAG, semantic search, or similarity tasks. Best for production embedding generation.

Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.

GPU-accelerated data curation for LLM training. Supports text/image/video/audio. Features fuzzy deduplication (16× faster), quality filtering (30+ heuristics), semantic deduplication, PII redaction, NSFW detection. Scales across GPUs with RAPIDS. Use for preparing high-quality training datasets, cleaning web data, or deduplicating large corpora.

Extract structured data from LLM responses with Pydantic validation, retry failed extractions automatically, parse complex JSON with type safety, and stream partial results with Instructor - battle-tested structured output library

Simplest distributed training API. 4 lines to add distributed support to any PyTorch script. Unified API for DeepSpeed/FSDP/Megatron/DDP. Automatic device placement, mixed precision (FP16/BF16/FP8). Interactive config, single launch command. HuggingFace ecosystem standard.

Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.

Fast tokenizers optimized for research and production. Rust-based implementation tokenizes 1GB in <20 seconds. Supports BPE, WordPiece, and Unigram algorithms. Train custom vocabularies, track alignments, handle padding/truncation. Integrates seamlessly with transformers. Use when you need high-performance tokenization or custom tokenizer training.

High-level PyTorch framework with Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks system, and minimal boilerplate. Scales from laptop to supercomputer with same code. Use when you want clean training loops with built-in best practices.

Applies cognitive science frameworks for creative thinking to CS and AI research ideation. Use when seeking genuinely novel research directions by leveraging combinatorial creativity, analogical reasoning, constraint manipulation, and other empirically grounded creative strategies.

---

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.

Framework for building LLM-powered applications with agents, chains, and RAG. Supports multiple providers (OpenAI, Anthropic, Google), 500+ integrations, ReAct agents, tool calling, memory management, and vector store retrieval. Use for building chatbots, question-answering systems, autonomous agents, or RAG applications. Best for rapid prototyping and production deployments.

Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.

Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.

Build complex AI systems with declarative programming, optimize prompts automatically, create modular RAG systems and agents with DSPy - Stanford NLP's framework for systematic LM programming

---

Expert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ models, 2/3/4/5/6/8-bit QLoRA, multimodal support

Language-independent tokenizer treating text as raw Unicode. Supports BPE and Unigram algorithms. Fast (50k sentences/sec), lightweight (6MB memory), deterministic vocabulary. Used by T5, ALBERT, XLNet, mBART. Train on raw text without pre-tokenization. Use when you need multilingual support, CJK languages, or reproducible tokenization.

Guarantee valid JSON/XML/code structure during generation, use Pydantic models for type-safe outputs, support local models (Transformers, vLLM), and maximize inference speed with Outlines - dottxt.ai's structured generation library

---

Expert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific model training

NVIDIA's runtime safety framework for LLM applications. Features jailbreak detection, input/output validation, fact-checking, hallucination detection, PII filtering, toxicity detection. Uses Colang 2.0 DSL for programmable rails. Production-ready, runs on T4 GPU.

Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task Arithmetic, linear merging, and production deployment strategies.

Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed checkpointing. Use when models exceed single-GPU memory or when you need DTensor-based sharding with DeviceMesh.

Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features. Use when discovering interpretable features, analyzing superposition, or studying monosemantic representations in language models.

Compress large language models using knowledge distillation from teacher to student models. Use when deploying smaller models with retained performance, transferring GPT-4 capabilities to open-source models, or reducing inference costs. Covers temperature scaling, soft targets, reverse KLD, logit distillation, and MiniLLM training strategies.

Use this skill when the user asks about hazardous materials (hazmat) transportation under 49 CFR 171-180 — placarding requirements, segregation tables, hazmat employee training (49 CFR 172 Subpart H), Hazmat Safety Permit (HMSP), shipping papers, packaging groups, emergency response phone, security plans, hazmat endorsement on CDL, and bulk vs non-bulk distinctions. Cite specific 49 CFR section.

Use this skill when the user asks how to pay CDL drivers — cents per mile (CPM), percentage of revenue, hourly, salary, sleeper-team rate, detention pay, driver-pay laws under FLSA (Fair Labor Standards Act), and how each model affects retention + recruitment. Cite FLSA + state wage laws.

Use this skill when a driver needs to move a CMV within a yard, terminal, or property under § 395.8 yard-move provisions. Covers what counts as yard move, ELD configuration, and where yard move ends and on-duty driving begins.

Use this skill when the user asks about state-specific trucking taxes BEYOND IFTA — NY Highway Use Tax (HUT), KY Kentucky Intrastate Tax (KIT), Oregon Weight-Mile Tax, New Mexico Weight-Distance Tax, Connecticut Highway Use Fee, or other state-level road-use taxes for commercial vehicles. Covers thresholds, rates, registration, filing cadence, and how these differ from IFTA. Reference state DOT/DOR resources.

Use this skill when the user asks about CDL driver medical certification — 49 CFR 391.41-49 medical qualifications, medical examiner's certificate (Long Form 5876 + Short Form 5875), National Registry of Certified Medical Examiners (NRCME), medical restrictions, federal disqualifying conditions, exemption programs (vision, diabetes, hearing), the FMCSA Medical Examiner's Handbook, or what to do when a driver's medical cert is downgraded. Cite specific 49 CFR 391 sections.

Use this skill when the user asks about logging / log hauling operations — short-haul exemption specifics for logging, log truck securement, axle weights, state-by-state logging permits (especially OR, WA, ID, BC), bunk + stake configurations, and how logging fits into FMCSA + state rules. Reference 49 CFR 393 for cargo securement of logs.

Use this skill to differentiate IFTA (fuel tax) and IRP (apportioned registration). Covers what each is for, deadlines, and how they interact.

Use this skill to understand what CSA data the public sees about your carrier and how that affects business. Covers SAFER, accessibility, and what's hidden vs visible.

Use this skill when the user asks about FMCSA financial assistance + grants — Motor Carrier Safety Assistance Program (MCSAP), High Priority Grants, CMV Operator Apprenticeship Program (CMV-DAP), workforce development funding, hazmat preparedness grants, small carrier financial relief, and state-level grants for trucking infrastructure. Reference fmcsa.dot.gov/grants.

Use this skill when conducting a self-audit of your DQF library before an FMCSA compliance review. Covers the systematic approach to verifying all 12 documents are current, complete, and properly maintained.

Use this skill when the user asks about workers compensation for trucking — required by state law, classification of drivers, premium calculation (rate × experience modifier × payroll), Common Claim types in trucking, returning injured drivers to work, claims management, premium reduction strategies, and the difference between WC and Occupational Accident insurance for owner-operators. Reference state WC laws + IFTA / fleet insurance.

Use this skill when a driver believes an inspection was conducted improperly. Covers how to request re-inspection, when state agencies will honor it, and what to expect.

Use this skill when selecting a Consortium/Third-Party Administrator (C/TPA) for drug + alcohol testing program management. Covers what services they provide, what to look for, and red flags.

Use this skill when the user asks about driver ergonomics — back/neck/shoulder injury prevention, seat adjustment + lumbar support, entering and exiting cab safely, load handling techniques, vibration reduction, stretching protocols, and how ergonomics affects workers compensation costs in trucking. Reference OSHA + ATA ergonomics research.

Use this skill when the user asks about truck purchase financing, leasing options, factoring, fuel surcharge math, lease-purchase agreements, the total-cost-of-ownership (TCO) comparison between buying + leasing, and how owner-operators finance their first truck. Cover Section 179 + Bonus Depreciation tax treatment.

Use this skill when the user asks about CDL classes (A, B, C), endorsements (H, N, P, S, T, X), required for hazmat / passenger / school bus / tank / triple-trailer operations, what each class is permitted to operate, the TSA Security Threat Assessment for hazmat, and ELDT (Entry-Level Driver Training) requirements. Cite 49 CFR 383.

Use this skill when the user asks about trucking rate negotiation — cost-plus pricing, market rate benchmarking via DAT/Truckstop, fuel surcharge math, how to negotiate with brokers/shippers, when to walk away, contract pricing vs spot rates. Reference DAT Rate Analytics + freight market reports.

Use this skill when managing post-SAP follow-up testing for a driver. Covers the testing cadence, recordkeeping, what happens if a test is missed, and how to transition the schedule to a new employer.

Use this skill when the user asks how to design or improve a fleet safety scorecard — what KPIs to track, how to weight them, telematics signals vs document compliance vs roadside outcomes, monthly cadence, driver-facing vs admin-facing dashboards, gamification considerations, peer-comparison metrics, and how a scorecard ties to insurance premiums + retention. Reference SMS Methodology v3.20 for industry-standard weights.

Use this skill when selecting a TPR-registered Training Provider for behind-the-wheel ELDT training. Covers the registry, evaluation criteria, and common pitfalls.

Audit, write, or revise Architecture Decision Records (ADRs), including the decision of whether one is warranted. Use when an architecturally significant decision needs recording, when reviewing an existing ADR collection, or when adding ADRs to a project that has none. Detects existing template (Nygard or MADR) and defaults to Nygard for new repos. Gates comparison ADRs through a decompose-and-verify interview before drafting.

Diátaxis