Top Agent Skills

Track ML experiments with open-source run logging, local or self-hosted dashboards, and media visualization using SwanLab.

Fine-tune large language models by training less than 1% of parameters using LoRA, QLoRA, and 25+ adapter methods, enabling efficient adaptation on consumer GPUs.

Quantize large language models to 4-bit with minimal accuracy loss using GPTQ, enabling deployment of 70B+ models on consumer GPUs with 4× memory reduction and 3-4× faster inference.

Connects LLMs with user data for RAG applications, document Q&A, and knowledge retrieval using 300+ data connectors and vector indices.

Generate image captions, answer visual questions, and perform image-text retrieval using BLIP-2's Q-Former architecture with frozen vision encoders and LLMs.

Train language models with SimPO, a reference-free preference optimization method that outperforms DPO without needing a reference model.

Post-train LLMs with reinforcement learning using the slime framework, which integrates Megatron-LM for training and SGLang for rollout generation.

Train large language models (7B-70B+) with RLHF using PPO, GRPO, DPO, and other algorithms, accelerated by Ray and vLLM for distributed multi-GPU setups.

Use RWKV, a linear-time RNN-Transformer hybrid, for efficient long-context inference and training with constant memory usage.

Train and run Mamba state-space models with O(n) complexity, achieving faster inference and longer context than Transformers.

Serve LLMs and VLMs with structured outputs, prefix caching, and high throughput using RadixAttention.

Quantize LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss using bitsandbytes. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers.

Generate high-quality sentence and text embeddings for semantic similarity, clustering, and retrieval using 5000+ pre-trained models. Supports multilingual and domain-specific embeddings for RAG and semantic search.

Run 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.

Segment any object in images using points, boxes, or masks as prompts, or automatically generate all object masks with zero-shot transfer.

Generate images from text prompts, perform image-to-image translation, inpainting, and build custom diffusion pipelines using Stable Diffusion models via HuggingFace Diffusers.

Optimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x memory reduction. Supports PyTorch native SDPA, flash-attn library, H100 FP8, and sliding window attention.

Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency on NVIDIA GPUs (A100/H100).

Add distributed training support to any PyTorch script with minimal code changes using a unified API for DDP, DeepSpeed, FSDP, and mixed precision.

Extend context windows of transformer models using RoPE, YaRN, ALiBi, and position interpolation techniques for processing long documents and implementing efficient positional encodings.

Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies for maximum GPU efficiency.

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.

Perform causal interventions on PyTorch models using pyvene's declarative framework for causal tracing, activation patching, and interchange intervention training.

Train and analyze Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features for mechanistic interpretability research.

Compiles research inputs—PDFs, code, logs, notes—into structured Agent-Native Research Artifacts with cognitive and physical layers.

Write Python code in n8n Code nodes, covering data access patterns, return formats, and standard library limitations.

Interpret n8n validation errors and warnings, guide iterative fixes, and explain validation profiles, error types, and auto-sanitization.

Detects Follow-Through Day (FTD) signals for market bottom confirmation using William O'Neil's methodology, with dual-index tracking and a state machine for rally attempt, FTD qualification, and post-FTD health monitoring.

Detects and ranks trending market themes by analyzing cross-sector momentum, volume, and breadth signals with lifecycle maturity assessment.

Analyzes market breadth using Monty's Uptrend Ratio Dashboard data to diagnose the current market environment. Generates a 0-100 composite score from 5 components (breadth, sector participation, rotation, momentum, historical context).

Convert daily market observations and news reactions into structured edge hints, with optional LLM augmentation, outputting a canonical hints.yaml for downstream concept synthesis.

Quantifies market top probability using O'Neil Distribution Days, Minervini leading stock deterioration, and Monty defensive sector rotation, producing a 0-100 composite score with risk zone classification for tactical 2-8 week timing signals.

Validate financial data quality in market analysis documents before publication, checking price scales, instrument notation, date accuracy, allocation totals, and unit usage.

Scans US equities for daily +20%/-20% movers, classifies catalysts and chart context, updates forward outcomes, and summarizes cohort patterns for research.

Generate Minervini-style breakout trade plans from VCP screener output with risk calculation, portfolio heat management, and Alpaca-compatible order templates.

Reviews trading strategy drafts for edge plausibility, overfitting risk, sample size adequacy, and execution realism, outputting PASS/REVISE/REJECT verdicts with confidence scores.

Screens recent post-earnings stocks using a 5-factor scoring system (Gap Size, Pre-Earnings Trend, Volume Trend, MA200 Position, MA50 Position) and assigns A/B/C/D grades to identify momentum trade candidates.

Quantifies market breadth health using public CSV data, generating a 0-100 composite score across six components to assess market participation and rally breadth.

Reviews closed trades, partial exits, and monthly trade aggregates for process adherence, risk discipline, execution quality, and evidence-based trading behavior patterns.

Evaluates account-level drawdown and losing-streak rules from local thesis files to decide whether new trade risk is allowed, without external APIs.

Review AI agent skills using a dual-axis method: deterministic code-based checks and LLM deep review, with weighted scoring and improvement recommendations.

Analyze and simulate options trading strategies using Black-Scholes pricing, Greeks calculation, and risk management guidance.

Validate multi-skill workflows by checking skill existence, inter-skill data contracts, JSON schema compatibility, file naming conventions, and handoff integrity.

Generate falsifiable trade strategy hypotheses from market data, trade logs, and journal snippets, with ranked hypothesis cards, experiment designs, kill criteria, and optional strategy.yaml export.

Routes trading and investing goals to the right workflow, skillset, API profile, and setup path without executing trades.

Evaluates a local pre-trade checklist before manual order entry, blocking planless, oversized, revenge-risk, market-regime-blocked, or circuit-breaker-blocked entries while journaling the decision for later review.

Aggregate closed trades from trader-memory-core into a weekly performance report with win rate, expectancy, profit factor, R-multiple, MAE/MFE, and pattern breakdowns by source skill, exit reason, sector, and mechanism.

Coordinate multi-stage edge research pipelines from candidate detection through strategy design, review, revision, and export.

Analyze historical downtrend durations and generate interactive HTML histograms showing typical correction lengths by sector and market cap.

Detect IBD-style Distribution Days for QQQ/SPY, track 25-session expiration and 5% invalidation, count d5/d15/d25 clusters, classify market risk, and emit TQQQ/QQQ exposure recommendations.

Synthesizes outputs from 8 upstream market analysis skills into a unified conviction score (0-100), pattern classification, and allocation recommendation for Druckenmiller-style portfolio positioning.

Screen US stocks for Stockbee-style short-term momentum burst setups using 4% breakout, dollar breakout, range expansion, volume expansion, prior range contraction, close-location, failure filters, and risk-distance scoring.

Classify US dividend tax treatment, validate holding periods, and recommend account placement for dividend income portfolios.

Screen US stocks for Stockbee-style selling-exhaustion hammer setups using prior momentum, pullback depth, undercut/reclaim, long lower-wick geometry, close-location, volume confirmation, quality/liquidity gates, and risk-distance scoring.

Generate AI images, videos, music, and audio from the terminal via muapi.ai — supports 100+ models including Flux, Midjourney v7, Kling 3.0, Veo3, and Suno V5.

Generate platform-tuned vertical short clips (YouTube Shorts, TikTok, Reels) from long-form videos using muapi.ai's AI clipping endpoint.

Generate a pack of professional or aesthetic photos from a single reference image while preserving the exact identity of the person.

Search the web and fetch content from URLs, returning LLM-friendly markdown.

Sets up an isolated Python environment for SageMaker and AWS work, pinning Python version and installing current boto3/awscli to avoid dependency conflicts and stale SDKs.

Creates SageMaker endpoints (real-time or async) with autoscaling, CloudWatch alarms, and tagging enabled by default, then smoke-tests them before declaring success.