LoopOS is a lightweight, 100% native autonomous loop protocol for your AI coding assistants (Gemini, Claude, OpenCode, and more).
It gives your agent a persistent VFS memory and structured execution pipeline inside a local .loop/ folder—allowing it to execute complex, multi-goal engineering tasks autonomously without external daemons, proxies, or background servers.
- 100% Native & Zero-Daemon: Runs entirely via standard Markdown files and your agent's native tools. No background Node servers or proxy daemons required.
- Stage-Driven Memory (
CONTEXT_MAP.md): Keeps prompt memory lean (Max 10 files per active stage) to prevent token window degradation on large codebases. - Topological Goal Graphs (
Depends On): Supports multi-goal projects with strict dependency graphs (e.g. Goal 2 waits for Goal 1). - Strict TDD Enforcement: Automatically writes failing unit tests (RED state) before writing implementation logic (GREEN state).
- Cross-Goal Regression Sweeps: Automatically re-runs legacy verification tests when completing new goals to ensure zero regressions.
- Pattern Consolidation (
PATTERNS.md): Distills codebase gotchas into a living, reusable design guide for future sessions. - Safe Surgical Rollbacks: Reverts targeted files safely via native git checkouts with full user permission visibility.
Install LoopOS globally across all your CLI agents with a single command:
npx loopos-install(Or clone the repository and run node install.js)
This activates the /loop protocol globally in Gemini CLI, Claude CLI, and OpenCode CLI.
In any project directory, ask your AI agent to run a loop:
> @agent "Use /loop Goal 1: Build User Auth in auth.js with tests. Goal 2: Build Profile API in profile.js with tests (Depends On: G1)."The agent will automatically scaffold a .loop/ folder, map out a topological pipeline, write an execution checklist, and execute the task autonomously until all goals pass verification!
LoopOS maintains state in a human-readable .loop/ directory inside your workspace:
GOALS.md— The goal ledger with topological dependency graphs.WORKFLOWS.md— High-level stage pipeline.PLANS.md— Active stage to-do list (automatically archived upon stage completion).CONTEXT_MAP.md— Active stage file bounds (Max 10 target files).VERIFICATIONS.md— Append-only verification command transcripts.DECISIONS.md— Architectural decision records & hypothesis pivots.PATTERNS.md— Consolidated codebase coding standards.
Apache License 2.0 — see LICENSE for details.