tack

# tack

[![npm version](https://img.shields.io/npm/v/tack-cli.svg)](https://www.npmjs.com/package/tack-cli) [![license: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)

Tack gives your agents a shared memory for the repo, so each new session can pick up where the last one left off.

Static instruction files go out of date as soon as the code changes. Tack keeps a shared record in `./.tack/` and checks that memory against the codebase, so the next agent starts with context that still matches reality.

## Install

1. Install Tack. If you prefer not to install globally, use `npx tack-cli@latest` instead of `tack`.

```bash
npm install -g tack-cli
```

2. Initialize the repo and install your agent startup instructions once.

```bash
tack init
tack setup-agent
```

3. Keep `tack watch` open in a second terminal so you can see what the agent is doing.

```bash
tack watch
```

4. Start the MCP server from the repo root with a visible agent label.

```bash
TACK_AGENT_NAME=cursor tack mcp
```

On Windows PowerShell, set the label like this instead:

```powershell
$env:TACK_AGENT_NAME="cursor"
tack.cmd mcp
```

5. Check `tack watch` for these events:

- `READY` means the MCP session connected
- `READ` means the agent read `tack://session` or another Tack context resource
- `WRITE` means the agent wrote memory back with `checkpoint_work`, `log_decision`, or `log_agent_note`

If you see `READY`, `READ`, and `WRITE`, the agent is actually using Tack.

When you wrap up a session, create a handoff:

```bash
tack handoff
```

Read the full guides:

- [Product Direction](./docs/product-direction.md)
- [Getting Started](./docs/getting-started.md)
- [Agent Workflow](./docs/agent-workflow.md)
- [MCP Clients](./docs/mcp-clients.md)
- [Watch Mode](./docs/watch.md)
- [CLI Reference](./docs/cli.md)
- [Detectors And YAML Rules](./docs/detectors.md)

## What Tack Actually Does

- Keep implementation facts aligned with the real codebase
- Stop stale project instructions from