Install Agent Context System in any project manually

Three ways to install Agent Context System: copy files into an existing project, create from the GitHub template, or clone as an OpenClaw workspace skill.

The Agent Context System is a set of files — no package manager or plugin required. You can add it to any project by copying the files directly, start a new project from the GitHub template, or install it as an OpenClaw workspace skill. All three approaches give you the same two-file system: AGENTS.md for committed project knowledge and .agents.local.md for your personal scratchpad.

Choose your install method

Existing project

Copy the Agent Context System files into a project you already have.

1. Clone and copy the files

   Run the following from your project root:

   git clone https://github.com/AndreaGriffiths11/agent-context-system.git /tmp/acs
   cp /tmp/acs/AGENTS.md /tmp/acs/agent-context ./
   cp -r /tmp/acs/agent_docs /tmp/acs/scripts ./
   rm -rf /tmp/acs
   

   This copies AGENTS.md, the agent-context CLI, the agent_docs/ reference directory, and the scripts/ directory into your project root, then removes the temporary clone.

2. Run init

   ./agent-context init
   

   The init command creates .agents.local.md from the template, adds it to .gitignore, and asks which agent tools you use. It then creates the appropriate config file for each tool you select.

3. Edit AGENTS.md

   Open AGENTS.md and replace the placeholder content with real information about your project: name, stack, commands, patterns, boundaries, and gotchas. This is the highest-leverage edit you will make.

New template project

Create a new repository directly from the Agent Context System template on GitHub.

1. Create the repository from the template

   gh repo create my-project --template AndreaGriffiths11/agent-context-system
   cd my-project
   

   Replace my-project with your repository name. This creates a new GitHub repository with all the Agent Context System files already in place.

2. Run init

   ./agent-context init
   

   Init sets up .agents.local.md, wires up the agent tool configs for whichever tools you select, and ensures the scratchpad is gitignored.

3. Edit AGENTS.md

   Open AGENTS.md and fill in your project-specific details. The template contains placeholder content — replace it with what is actually true about your project.

Starting from the template is the fastest path to a working setup. You get AGENTS.md, the agent-context CLI, all scripts, and the agent_docs/ directory structure in one step.

OpenClaw

OpenClaw reads workspace skills directly from a skills/ directory. No init command is needed.

1. Clone into your skills directory

   From your project root:

   git clone https://github.com/AndreaGriffiths11/agent-context-system.git skills/agent-context-system
   
2. Restart your OpenClaw session

   OpenClaw picks up workspace skills on session start. Restart your session after cloning, and OpenClaw will read AGENTS.md automatically.

3. Edit AGENTS.md

   Open skills/agent-context-system/AGENTS.md and fill in your project specifics.

Unlike the other install methods, OpenClaw does not require running ./agent-context init. OpenClaw’s skill system handles loading the context automatically.

What init creates

After running ./agent-context init, your project will contain the following files:

File	Purpose
AGENTS.md	Committed project knowledge. Always loaded. Keep under 120 lines.
.agents.local.md	Gitignored personal scratchpad. Grows as the agent logs session learnings.
CLAUDE.md	Symlink to AGENTS.md. Created if you selected Claude Code during init.
.cursorrules	One-line directive for Cursor. Created if you selected Cursor during init.
.windsurfrules	One-line directive for Windsurf. Created if you selected Windsurf during init.
.github/copilot-instructions.md	Directive for GitHub Copilot. Created if you selected Copilot during init.
agent_docs/	Deeper reference docs for your agent to load on demand.
scripts/	Maintenance scripts for init, validation, promotion, and compression.

.agents.local.md is added to .gitignore automatically. It is personal and should not be committed.

Validate your setup

After init, run the following to confirm everything is wired up correctly:

agent-context validate

The validate command checks that AGENTS.md exists and is under 120 lines, that .agents.local.md is gitignored, and that at least one agent tool config is present.

Other CLI commands

agent-context init                  # Set up context system in current project
agent-context validate              # Check setup is correct
agent-context promote               # Find patterns ready to move to AGENTS.md
agent-context promote --autopromote # Auto-append patterns recurring 3+ times

After setup

Once init completes:

1. Fill in AGENTS.md with your actual project name, stack, commands, and the patterns and gotchas specific to your codebase.
2. Fill in agent_docs/ with deeper references your agent should load on demand — architecture decisions, full conventions, extended gotchas. Delete any sections that do not apply.
3. Customize .agents.local.md with your personal preferences (planning style, code preferences, tone).
4. Work normally. Your agent reads AGENTS.md and .agents.local.md at session start, does the task, and updates the scratchpad at session end.
5. Promote what sticks. When a pattern recurs across three or more sessions, the agent flags it in the scratchpad’s “Ready to Promote” section. You decide when to move it into AGENTS.md.