Drifter — AI skill for Claude Code

Drifter — Universal AI Agent Drift Guard

**Before you write a single line of code, know what already exists.**

AI coding agents excel at local optimization and fail at project discipline over long horizons. Drifter is a lightweight governance layer — not a prompt, not a replacement — that keeps agents aligned with project structure as codebases evolve.

• Document type system — every doc knows its mutation rules; docs don't become lies
• Project conductor — one active task at a time; scope creep is structurally prevented
• 35 automated drift checks — stale refs, doc drift, credential leaks, dead code, file-size bloat, ghost references, and more
• Pre-flight checklist — 7 designed-to-be-followed steps before any code change
• Command boundaries — agents read dangerous_patterns.toml before running shell
• Real enforcement — ShellGuard.enforce() raises DangerousCommandError or ApprovalRequiredError on violation
• Plugin API — ToolInterceptor auto-logs and enforces before any tool call (MCP, Kimi, custom)
• Cross-platform shell history — audit bash, zsh, and fish history with auto-detection
• Git pre-commit hook — block commits that introduce drift (drifter install-hook)
• Session audit — HMAC-signed tamper-evident log of reads, writes, and checks; forged entries are detected and skipped
• Granular suppression — per-check ignore_paths and ignore_patterns via drifter.toml

Drifter is a cooperative guard — it raises the cost of mistakes and makes them auditable, not impossible. For mandatory enforcement, pair it with git pre-commit hooks and CI gates.

Drifter is not a prompt. It is a **system of enforced protocols**.

Quick Start

Requires **Python 3.10+**.

# Install
pip install drifter

# Initialize in your project
cd your-project
drifter init

# Run the drift guard
drifter check

# Run pre-flight before a coding session
drifter preflight --task "fix login bug"

What You Get

| Artifa