The outcome

Move from a bounded request to a verified diff while preserving context, ownership, and a clean review trail.

Amp is a cloud-connected AI coding agent for terminal and editor workflows. Running focused coding tasks in durable threads, applying repository guidance, reviewing changes, and using different reasoning modes from the terminal or editor. 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

    Start Amp in the intended repository on a clean branch and confirm the active project, thread visibility, status, and repository guidance files.

  2. 02

    Create one thread for the task and state the outcome, acceptance criteria, relevant paths, constraints, non-goals, and validation commands.

  3. 03

    Choose low, medium, high, or ultra mode according to task difficulty, then ask Amp to inspect and explain its plan before editing.

  4. 04

    Review tool calls, commands, files, dependencies, and intermediate tests, correcting scope immediately instead of accumulating unrelated work.

  5. 05

    Run trusted checks and inspect the final diff yourself, then use the approved branch, pull-request, or ship workflow only after human review.

Working standard

What good use looks like.

  • Use one thread per task.
  • Put durable rules in AGENTS.md.
  • Match reasoning mode to task difficulty.

Amp runs tools without asking by default, and thread data can include prompts, tool results, attachments, and selected code. Work in an isolated branch or environment, add custom permission controls for untrusted inputs, keep secrets out of context despite best-effort redaction, choose thread visibility deliberately, review workspace and administrator access, inspect every diff and command, and require human approval for pushes, deployments, migrations, or destructive actions.

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.