The outcome

Use Roo's specialized modes without giving an agent broader access or autonomy than the task requires.

Roo Code is a open-source AI coding agent extension for VS Code. Planning, implementing, debugging, and explaining repository changes with specialized modes and visible approval of file, command, browser, and MCP actions. This guide narrows that broad capability into one repeatable outcome, with checkpoints that keep the source material and your judgment in the loop.

Before you begin

Set the boundary before the tool starts.

Choose one real task, identify who will use the result, and decide what evidence or test will make the result acceptable. Gather only the source material needed for that task. If the work contains confidential, personal, regulated, or client-owned information, confirm that the platform and account are approved before sharing it.

Troiana principle

AI should make the work easier to inspect. If the workflow removes the source, the owner, or the review step, redesign the workflow.

Step by step

A workflow you can repeat.

  1. 01

    Create a clean branch, capture the baseline status and tests, and inspect repository rules, architecture, secrets, and generated-file boundaries.

  2. 02

    Choose Ask or Architect mode first and provide the outcome, acceptance criteria, relevant paths, non-goals, and exact validation commands.

  3. 03

    Review Roo's explanation and minimal plan, then switch deliberately to Code mode with auto-approval disabled or narrowly allowlisted.

  4. 04

    Approve only understood tool calls, inspect edits and terminal commands as they occur, and reject browser, MCP, dependency, or scope expansion without justification.

  5. 05

    Run trusted checks independently, inspect the complete diff and cost, and submit through normal peer review without granting Roo release authority.

Working standard

What good use looks like.

  • Plan before Code mode.
  • Keep command auto-approval narrow.
  • Verify the final diff independently.

Roo can read and edit files, execute terminal commands, browse, call MCP servers, switch modes, and create subtasks. Keep auto-approval disabled or narrowly allowlisted, treat repository and web content as untrusted, isolate consequential work, restrict MCP tools and credentials, review cost limits and provider data handling, inspect every diff and command, and never delegate merges, deployments, migrations, or destructive actions without explicit authorization.

Official references

Check the current product documentation.

Features, plan limits, availability, and data controls change. These official pages are the starting points used for this collection.