Context Layer — Human-in-the-loop Context Engineering for AI-Assisted Development

2 min read Original article ↗

Context engineering makes or breaks your agents

Models have no persistent memory across sessions.

Same clarifications, same architecture, every time.

Redundant context burns the attention budget.

Decisions and intent never persist for the model.

As tokens grow, recall drops — context rot.

Automated context management clutters the context. Curation needs a human.

Further reading: Anthropic on effective context engineering · Chroma on context rot

Meet Context Layer — curated domain intelligence for AI agents

Simple

No magic, just folders and symlinks.

Reusable

Reuse the context between projects.

Clean

Doesn't pollute your repo. Context is stored separately from domain code.

Agent-agnostic

Cursor, Claude, future agents.

Persistent

Store the context on your file system or in a dedicated GitHub repository.

Human-in-the-Loop

You curate your domain context.

Built for engineers who treat context as a first-class resource

AI-first developers Independent builders using LLMs daily Engineers automating repeatable work Developers running multi-agent workflows Founders optimizing AI development cost

How it works

This is how your project directory looks like with the context layer.

my-project/
├── README.md
├── ...  # your project files
└── .ctxlayer/  # symlink to ~/.agents/ctxlayer/
    └── my-domain/
        ├── config.yaml
        ├── task-1/
        │   ├── docs/
        │   │   ├── 01-initial-research.md
        │   │   └── 02-implementation-notes.md
        │   └── data/
        │       └── sample-data.json
        └── task-2/
            ├── docs/
            └── data/
  • The context layer is a directory structure on your machine.
  • Reference domain intelligence across domains via symlinks.
  • The context layer is not stored in the project repo, but can be persisted in a dedicated GitHub repo.
  • Reference the context from agent prompts by saying "in the context layer".

Quick start

  1. Install the ctx CLI and agent skill.
  2. From your project directory, run ctx new — Create a new task to hold the context for the task you are about to do. Initializes the context layer for your domain if it was not yet initialized.
  3. Add docs/data — Add your documentation and reference material to the task's docs/ and data/ folders.
  4. Ask agent — From your prompts, say "in the context layer, …". This keyword is understood by the agent and refers to the context layer directory of the active task.

Continue to the Docs →