OneShotDev

A VS Code extension: one prompt → AI writes code → runs your tests → fixes failures in a loop → deploys the result to AWS, all from inside the editor.

Supports Claude, OpenAI, and Gemini — pick one as your active provider (Claude is default), you supply each key you want available.

What it actually does

1. You type one prompt describing what to build.
2. The selected AI generates a full project (code + tests + manifest) and writes it into your open workspace folder.
3. It runs your configured test command (default npm test).
4. If tests fail, the error output is fed back to the AI to fix, and it retries — up to oneshotdev.maxFixAttempts times (default 4).
5. Once tests pass, it asks if you want to deploy. If yes: builds a Docker image, pushes it to AWS ECR, and creates/updates an AWS App Runner service running it.

Why App Runner: it's the simplest AWS option for "deploy a container and get a public HTTPS URL" — no load balancer, VPC, or cluster config needed (unlike ECS/Fargate or EC2).

Install (development mode)

npm install
npm run build

Then in VS Code: Run > Start Debugging (or press F5) from this folder — that launches an Extension Development Host with OneShotDev loaded.

To package it properly for normal install:

npm install -g @vscode/vsce
vsce package

This produces a .vsix you can install via Extensions → ... → Install from VSIX.

Publish to the VS Code Marketplace (make it public)

Publisher ID for this project: mayurmj1994

1. Get a Personal Access Token from https://dev.azure.com → User settings → Personal Access Tokens → scope: Marketplace (Manage).
2. Push this code to your repo: https://github.com/mayurmj2023/oneshotdev (the repository.url in package.json is already set to this.)
3. Run:

   npm install -g @vscode/vsce
   vsce login mayurmj1994
   vsce publish
   

4. It goes live at https://marketplace.visualstudio.com/items?itemName=mayurmj1994.oneshotdev within a few minutes — anyone can install it from the Extensions panel in VS Code.

To update later: bump "version" in package.json, then vsce publish again.

Configure (Settings → search "oneshotdev")

One-time AWS setup (required before first deploy)

App Runner needs an IAM role that lets it pull images from your private ECR repo. Create it once:

aws iam create-role --role-name AppRunnerECRAccessRole \
  --assume-role-policy-document '{
    "Version": "2012-10-17",
    "Statement": [{
      "Effect": "Allow",
      "Principal": { "Service": "build.apprunner.amazonaws.com" },
      "Action": "sts:AssumeRole"
    }]
  }'

aws iam attach-role-policy --role-name AppRunnerECRAccessRole \
  --policy-arn arn:aws:iam::aws:policy/service-role/AWSAppRunnerServicePolicyForECRAccess

Copy the resulting Role ARN into oneshotdev.ecrAccessRoleArn.

You also need, on the machine running VS Code:

• Docker installed and running
• AWS CLI credentials configured (aws configure, or env vars, or an AWS SSO profile) with permissions for ECR + App Runner + STS

Cost & safety notes

• Each generate/fix call costs you API credits on whichever provider you pick.
• App Runner bills per vCPU/memory while running — it's not serverless-free like Lambda. For light/dev workloads this is usually a few dollars/month, but keep an eye on it.
• The extension only writes files inside your open workspace folder and refuses any path that tries to escape it (no ../ traversal).
• Generated code is never eval'd — it's written to disk and run via your own test command, same as code you'd write yourself.