Loop Engineering Language

Write .loop files — structured task contracts that tell AI agents exactly what to build, what tools they can use, and how to confirm it worked.

Unlike a chat prompt, a .loop file is a compiler-checked specification. Wrong brackets cause a syntax error. Missing required blocks are caught before the agent starts.

Features

• Syntax highlighting — block keywords, tool declarations, field names, types, and return arrows all highlighted distinctly
• Snippets — type goal, task, loop, etc. to expand full blocks with tab stops
• Bracket enforcement — Goal uses [ ], everything else uses { }, wrong bracket = error
• File icon — .loop files get their own icon in the Explorer

Quick look

Goal [
    Build a REST API with JWT authentication and PostgreSQL persistence.
]

Task {
    "Create users table migration"
    "Implement POST /auth/register and POST /auth/login"
    "Protect GET /api/tasks with auth middleware"
}

Discovery {
    scan: ["src/**/*.rs", "migrations/**/*.sql"]
    find: [
        "Does an auth module already exist?"
        "What database connection setup is in place?"
    ]
}

Planning {
    steps: [
        "Run discovery"
        "Create database schema"
        "Implement auth endpoints"
        "Write integration tests"
    ]
    max_iterations: 6
}

Execution {
    tools: [
        read_file(path: string) -> string
        write_file(path: string, content: string) -> bool
        run_command(cmd: string) -> string
    ]
    strategy: "Execute steps in order. Read before writing. Verify after each step."
}

Verification {
    checks: [
        "cargo test passes with exit code 0"
        "POST /auth/login returns 200 with a JWT"
        "GET /api/tasks returns 401 without Authorization header"
    ]
    on_fail: retry
    max_retries: 4
}

The seven blocks

Snippets

Type the start of a block name and press Tab:

Tab through each placeholder to fill in the details.

Bracket rules

Goal is the only block that uses square brackets. Everything else uses curly braces. Tool parameters use parentheses. The compiler rejects wrong brackets with a clear message:

// Correct
Goal [ describe the task here ]
Memory { project_path: "./myapp" }
Task { "do this" "then this" }

// Syntax errors
Goal { describe the task }   // ERROR: Goal requires [ ] not { }
Task [ "item" ]              // ERROR: Task requires { } not [ ]
Execution ( ... )            // ERROR: Execution requires { } not ( )

CLI

The extension pairs with the loop CLI for validation and execution:

npm install -g @squareexp/loop   # install

loop init                         # create .loop/skills, Memory/, Goal.loop
loop check Goal.loop              # validate the file — shows per-block status
loop run Goal.loop                # run with an AI agent
loop verify Goal.loop             # run verification checks
loop status                       # show current state and failure history
loop inspect Goal.loop            # print the parsed AST

Install: npmjs.com/package/@squareexp/loop

How it works

1. Write a .loop file describing what you want built
2. Run loop init — creates the workspace structure and drops a skill document that explains the language to the agent
3. Give the .loop file to an AI agent (Claude, Gemini, GPT-4, etc.)
4. The agent reads Discovery, runs Planning steps with the declared tools, and checks Verification after each iteration
5. Run loop verify Goal.loop when you think it's done — if all checks pass, state is updated and the file is marked complete

Every failed tool call and failed check is recorded in .loop/state.json. The agent reads this on each iteration and doesn't repeat the same mistakes.

Requirements

• VS Code 1.60 or later
• loop CLI for running and validating files (optional but recommended)

Source

github.com/squareexp/loop