Context engineering

The model only knows what is in front of it. Context engineering is deciding, deliberately, what that is: the material it needs, the constraints it must respect, examples of good output, and the task itself, in that order, and nothing it does not need. It replaces the hope that one clever sentence will do the work.

Standing: Rising since 2025-06
Stars, week: +3,252 across 6 repositories
Stars, month: +22,838, through 2026-11-08
Listings: 6 chosen by hand

The standing

Where it stands

Among the 14 concepts. Bars: GitHub stars gained this week.

Established · 7

Rising · 7

How to use it today

In ChatGPT or Claude, as a person doing the work, not as a developer.

1Start a new chat for each piece of work, so old turns do not crowd out what matters now.
2Paste or attach the material first (the notes, the data, the brief), then write the constraints (audience, length, format, what to leave out), then the task, as separate labeled blocks.
3Give one example of the output you want, even a rough one; the model copies shape far more reliably than it follows adjectives.
4In ChatGPT use Projects, in Claude use Projects or a CLAUDE.md file in Claude Code, to hold the standing material so you stop re-pasting it.
5When the answer is off, fix the context (what was missing or contradictory), not the wording of the request.

What changes day to day

You stop writing long requests and start assembling short folders of material. The first message gets longer and more structured; the follow-ups get shorter. Output quality stops depending on your phrasing.

Where it goes wrong

More context is not better context.

Stuffing everything in makes the model miss the one thing that mattered. Contradictory instructions across pasted documents are the commonest cause of odd answers.

Listings that use it well

Chosen by hand after reading them, each with the reason. Install from the listing's own page.

Listing	Kind	Why it is here	Stars
agent-skills	Agent skills	Ships a dedicated context-engineering skill that configures rules files and project context and is re-run when output quality degrades.	114k
Context7	MCP server	Pulls version specific library documentation into the prompt on demand instead of relying on what the model remembers.	64k
ai-agent-book	Agent skills	Book chapters packaged as skills on context engineering (static prefix plus trajectory, system prompt structure, progressive disclosure of skills) and on context compression.	57k
claude-howto	Agent skills	Its claude-md skill creates and audits CLAUDE.md files, the standard project context file for Claude Code.	47k
Repomix	MCP server	Packs a whole repository into one AI friendly file so a codebase can be handed to a model as structured context.	29k
planning-with-files	Agent skills	Keeps task_plan, findings and progress files on disk and re-injects them each turn to survive /clear, compaction and context rot.	28k

Elsewhere

Repositories that embody the idea and are not in the catalog.

davidkimai/Context-Engineering ↗First-principles handbook on context design, orchestration and optimization beyond prompts.
humanlayer/12-factor-agents ↗Principles for production LLM software; several factors are about owning the context window.
coleam00/context-engineering-intro ↗Template for giving a coding assistant the context it needs (CLAUDE.md, PRPs, examples, validation).

Standing

Our judgment, with the reason and the sources it rests on.

Rising since 2025-06. Popularized in June 2025 by Shopify's Tobi Lütke and Andrej Karpathy, and given vendor guidance by Anthropic in September 2025. It now leads advice on building agents; in everyday use the older name, prompt engineering, is still more common. Adopted faster each month; the vendors are building for it and new tools keep appearing.

The rise of "context engineering"2025-06-23
Andrej Karpathy on X: +1 for "context engineering" over "prompt engineering"2025-06-25
Effective context engineering for AI agents2025-09-29
Tobi Lütke on X: I really like the term context engineering over prompt engineering2025-06-19