Drift lets your agent move fast and break less

You prompt an agent. It writes code fast — and quietly drifts away from what you asked for. Every edit adds a small assumption you never approved; over a long task, they pile up. Drift catches that.

Drift makes your agent:

• More accurate. It checks its own work against your rules and requirements, before telling you it's done.
• Fewer prompts. Say what you want once; the agent keeps it in mind across every edit after.
• Better memory. Your project's rules and requirements live in the repo and are statically linked to your code.
• Fewer mistakes. When a change breaks a rule, the agent finds out before you do.
• Pushback when it matters. Ask for something that contradicts your own requirements, and the agent flags it: "Are you sure? This goes against what you wanted earlier."
• Long tasks that stay reliable. Steps get checked much more often, so a 50-step task doesn't quietly pile up 50 small wrong turns.
• Work that outlives the conversation. Close the tab, start fresh, hand it to another agent — the requirements are still there, no re-explaining required.

No new workflow to learn

You keep prompting the way you already do. Drift is a tool your agent picks up, not one you have to learn. The agent writes the specs, places the markers, runs the checks, and resolves any drift — your job is to review.

Behind the scenes: specs are your rules and requirements in plain English, markers wrap the code that implements each requirement, and drift is the signal that fires when the two disagree. Start with Why drift exists for the full picture.

Quickstart

1. Install drift (see Install below).
2. Give the binary to your LLM agent and tell it to run drift skill.

That's it. The agent learns everything it needs from drift skill - how to write specs, place markers, link them, check for drift, and resolve it. You prompt; the agent does the rest.

Documentation

The full docs live in docs/. Start with Why drift exists for the problem and how drift solves it, or jump straight to Getting started to install and hand off to your agent.

Install

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/SteelSprint/Drift/main/scripts/install.sh | bash

Windows (PowerShell):

irm https://raw.githubusercontent.com/SteelSprint/Drift/main/scripts/install.ps1 | iex

Or pin a version:

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/SteelSprint/Drift/main/scripts/install.sh | DRIFT_VERSION=v1.0.0 bash

# Windows
$env:DRIFT_VERSION='v1.0.0'; irm https://raw.githubusercontent.com/SteelSprint/Drift/main/scripts/install.ps1 | iex

Installs to ~/.local/bin/drift (macOS/Linux) or %USERPROFILE%\.local\bin\drift.exe (Windows); override with DESTDIR. To build from source: make build.

Drift dogfoods itself.

Drift dogfoods itself. Every feature, every CLI command, every design decision in this repo has a spec. make build runs drift todo as a gate: if any spec drifted from its code, the build fails. No exceptions. Browse the specs.