karpathy-guidelines
Behavioral guidelines for LLMs to reduce coding mistakes, follow best practices, and improve output quality by enforcing simplicity, surgical changes, and goal-driven verification.
Discover reusable agent skills, browse implementation details, and find the right skill for your workflow.
542 skills found
Behavioral guidelines for LLMs to reduce coding mistakes, follow best practices, and improve output quality by enforcing simplicity, surgical changes, and goal-driven verification.
High-performance document intelligence library for extracting text, tables, code, and metadata from 91+ file formats, with OCR and LLM-ready output.
Stateful, multi-session workflow manager that uses Work Logs to persist context, enable checkpoints, and manage iterative progress across long-running tasks.
Statistical visualization library for Python. Create publication-quality graphics like box plots, heatmaps, and violin plots with pandas integration and automatic statistical estimation.
Official Pinia state management skill for Vue. Expert guidance on store definition, reactivity, type-safety, and store-to-store communication.
Evaluate Deca agent prompts and behavioral consistency through automated test runners, manual LLM judgment, and structured reporting.
Professional PDF generation from markdown using Pandoc. Supports Eisvogel templates, EB Garamond typography, and dual-layout optimization (A4 for print/desktop, 6x9 for mobile/Telegram).
Fixes CJS/ESM module compatibility issues in Nango integrations after zero-yaml migration, including path adjustments, ESM wrappers, and restoring original implementations.
Execute z.AI CLI for multimodal analysis, web search, reader, and GitHub repo exploration via CLI and MCP.
Build scalable, production-ready design systems using Tailwind CSS v4, featuring CSS-first configuration, design tokens, component variants, and responsive patterns.
Guided, systematic feature development agent that orchestrates codebase exploration, architectural design, implementation, and automated testing.
Anthropic Claude integration patterns: streaming, RAG with pgvector, tool use, model selection (Haiku/Sonnet/Opus), prompt caching, and cost management for AI-powered engineering.