Previous
0 / 3

Module questions

VC-00 · Orientation and Guardrails

Module VC-00VIBECODE

Create a Permission Plan

VIBECODE • Vibe Coding: AI-First Software Delivery

Browser-only practice

Problem Statement

Stage: Orientation and Guardrails

An AI agent proposes operations with very different risk levels.

Deliverable

A line-by-line permission decision.

Input and output contract

Read operation:target lines; mark read, search, and test as auto, and all other operations as approval.

Acceptance criteria

  1. Preserve target text.
  2. Use auto only for read-only or test actions.
  3. Require approval for unknown actions.

Suggested vibe loop

Read AGENTS.md, connect your Ollama or OpenAI-compatible provider in the browser, and ask the agent to restate the contract and make one small verifiable change. Watch edits stream into app.py while the editor is locked, then review the result and run the visible example. Keep the checkpoint when the evidence is correct or use Undo to restore it. The browser-client agent does not require an administrator gateway or API key.

Constraints

Use Python 3 standard library only. Read and follow AGENTS.md, keep main.py as the entry point, and implement solve(raw) in app.py.

Sample Testcases

Submit runs every public testcase in this browser. Results and code never leave this device.

Sample #1
Public sample
Input
read:src/app.py
write:src/app.py
test:unit
delete:data.db
Output
read:src/app.py -> auto
write:src/app.py -> approval
test:unit -> auto
delete:data.db -> approval
Explanation
Separates safe inspection from mutation.

Web terminal

C, C++, Java, and Python run locally in a browser VM. No worker or visualizer is used.

Saved in this browser
Editor settings