Claude Skills by anubhavg-icpl

Given a confusion matrix and class names, surface per-class failures and propose the single most impactful fix. Use when you need help with skill classification diagnostics.

Choose the right clustering algorithm based on data shape, noise, and constraints. Use when you need help with skill clustering guide.

Instrument a production VLM endpoint with Cross-Modal Error Rate monitoring, dashboards, and alerts. Use when you need help with skill cmer monitor.

Quick reference for complex number operations in ML and signal processing contexts. Use when you need help with skill complex arithmetic.

Turn user utterances into well-formed SAM 3 concept prompts with splitting, disambiguation, and fallbacks. Use when you need help with skill concept prompt designer.

Decision framework for designing context assembly pipelines based on task type, window size, and latency budget. Use when you need help with skill context engineering.

Walk a CNN spec layer by layer and report output shape, receptive field, and parameter count for every block. Use when you need help with skill conv shape calculator.

Determine if an optimization problem is convex and choose the right solver. Use when you need help with skill convexity checker.

Decision framework for LLM cost optimization -- caching strategies, rate limiting, model routing, and budget controls. Use when you need help with skill cost patterns.

Decision framework for choosing the right reasoning technique based on task complexity, accuracy requirements, and cost constraints. Use when you need help with skill cot patterns.

Write greedy and beam-search CTC decoders from scratch, including length normalisation. Use when you need help with skill ctc decoder.

Write a complete DCGAN scaffold from z_dim, image_size, and num_channels, including training loop and sample saver. Use when you need help with skill dcgan scaffold.

Decision-tree checklist for debugging neural network training failures. Use when you need help with skill debug checklist.

Build point clouds from depth maps with correct intrinsics handling and export to .ply. Use when you need help with skill depth to pointcloud.

Choose the right dimensionality reduction technique for a given task based on data size, goal, and downstream use. Use when you need help with skill dimensionality reduction.

Production patterns for embeddings, vector search, and similarity. Use when you need help with skill embedding patterns.

Choose the right ensemble method and configure it for your problem. Use when you need help with skill ensemble builder.

Decision framework for choosing evaluation strategies -- when to use which method, how to size test suites, and how to integrate evals into CI/CD. Use when you need help with skill eval patterns.

Evaluation strategy checklist for classification and regression models. Use when you need help with skill evaluation.

Quick reference decision tree for choosing the right feature selection method. Use when you need help with skill feature selector.

Decision tree for when and how to fine-tune LLMs with LoRA and QLoRA. Use when you need help with skill fine tuning guide.

Audit a video pipeline's frame sampler for off-by-one, short-clip handling, and crop consistency. Use when you need help with skill frame sampler auditor.

Report which parameters are trainable, which BatchNorm layers are in eval mode, and whether the optimizer is actually consuming the trainable parameters. Use when you need help with skill freeze inspector.

Decision framework for implementing function calling in production -- tool design, error handling, security, and provider patterns. Use when you need help with skill function calling patterns.

Compute gradients of common ML loss functions and choose the right derivative approach. Use when you need help with skill gradient computation.

Analyze graph-structured data and choose the right graph algorithm for ML tasks. Use when you need help with skill graph analysis.

Decision framework for choosing and implementing guardrails in production -- tool selection, layering strategy, and cost-performance tradeoffs. Use when you need help with skill guardrail patterns.

Write the sub-pixel heatmap-to-coordinate routine used by every production pose model. Use when you need help with skill heatmap to coords.

Inspect any image-shaped tensor or array and report dtype, layout, range, and whether it looks raw, normalized, or standardized. Use when you need help with skill image tensor inspector.

Build an image embedding index with any CLIP checkpoint; support query-by-text and query-by-image. Use when you need help with skill image text retriever.

Decision checklist for handling imbalanced classification problems. Use when you need help with skill imbalanced data.

Diagnose and optimize LLM inference serving throughput, latency, and cost. Use when you need help with skill inference optimization.

Apply information theory concepts to ML loss functions, model evaluation, and feature selection. Use when you need help with skill information theory.

Functional programming patterns in JAX -- when and how to use grad, jit, vmap, and pmap. Use when you need help with skill jax patterns.

Write a complete latency-benchmarking script with warmup, synchronisation, percentiles, and memory tracking. Use when you need help with skill latency profiler.

Generate a Voyager-shaped skill library with registration, retrieval by similarity, compositional execution, and failure-driven refinement. Use when you need help with skill library.

Write the complete linear-probe evaluation for any frozen encoder and labelled dataset. Use when you need help with skill linear probe runner.

Write a full LoRA training config for a custom dataset, including captions, rank, batch size, and learning rate. Use when you need help with skill lora training setup.

Generate the exact code for swapping box and mask heads on a torchvision Mask R-CNN for a custom num_classes. Use when you need help with skill mask rcnn head swapper.

Write a complete evaluation harness for MOTA / IDF1 / HOTA against ground-truth tracks. Use when you need help with skill mot evaluator.

Choose the right Naive Bayes variant for your classification task. Use when you need help with skill naive bayes chooser.

Produce a linear, cosine, or sigmoid beta schedule given T and target corruption level, plus SNR plot. Use when you need help with skill noise schedule designer.

Understand the perceptron pattern and when to use single-layer vs multi-layer architectures. Use when you need help with skill perceptron.

Automated checks for object permanence, gravity, and continuity on any generated video before shipping. Use when you need help with skill physical plausibility checks.

Given target latency and throughput, assign a time budget to every pipeline stage and flag which stage will miss its budget first. Use when you need help with skill pipeline budget planner.

Write a PyTorch Dataset for .ply / .pcd / .xyz files with correct normalisation, centring, and point sampling. Use when you need help with skill point cloud loader.

Choose the right probability distribution for a given ML problem. Use when you need help with skill probability reasoning.

Decision framework for shipping LLM applications to production -- covers every component with specific thresholds and pass/fail criteria. Use when you need help with skill production checklist.

Decision framework for choosing the right prompt pattern based on task type, reliability requirements, and target model. Use when you need help with skill prompt patterns.

Reference patterns for PyTorch training, evaluation, and deployment. Use when you need help with skill pytorch patterns.