Agent Skills for Context Engineering
ActiveOpen-source collection of transferable Agent Skills for building production AI agent systems on Claude Code, Cursor, or custom stacks.
Overview
A collection of Agent Skills for context engineering: the discipline of managing language model context windows so agents stay effective. Where prompt engineering optimizes individual prompts, context engineering covers everything that enters the model’s attention budget.
The repo packages transferable patterns for building production AI agent systems on Claude Code, Cursor, or custom implementations.
Skills
The library covers four areas:
- Foundations: context fundamentals, degradation patterns, compression strategies.
- Architectural patterns: multi-agent systems, memory architectures, tool design, filesystem-based context.
- Operations: context optimization, evaluation, LLM-as-judge.
- Cognitive models: BDI mental states for deliberative reasoning.
A recent addition is Long-Horizon Prompting: a method for writing launch briefs for autonomous runs that span hours, context windows, or parallel workers.
Production Examples
Example pipelines and sites that ship with the repo include a digital-brain skill, an LLM-as-judge evaluation harness, a book SFT pipeline that fine-tunes an 8B model on Gertrude Stein’s literary style for about $2 in compute, and the Long-Horizon Prompting skill with a companion Prompt Lab.
Recognition
- Reached #1 on Replicate Hype and passed 15,000 GitHub stars.
- Cited in “Meta Context Engineering via Agentic Skill Evolution” (2026, Peking University) on static skill architecture.
- Mapped in “Agent Harness Engineering: A Survey” (2026, OpenReview / TMLR submission) under context and working-state engineering; the paper also references it as a production skill library covering KV-cache compaction patterns between agent calls.
- Featured in TLDR AI and Ben’s Bites.