Get human code review and fixes for AI-generated projects

Sets up an external human code-review service (Vibers) that reviews AI-generated code pushed to a GitHub repository against a written spec, then submits a follow-up PR with fixes made directly in the code. Setup is three steps: add marsiandeployer as a repository collaborator; add a GitHub Action workflow, triggered on push to main, that points to a publicly-accessible spec document (Google Doc, Notion, or similar - review is impossible without spec access) and a Telegram contact for notifications, with an optional review_scope of full (default), security, or spec-compliance; and add a "How to test" requirement to the project's own agent instructions (CLAUDE.md, .cursorrules, or AGENTS.md) so every commit message includes a live URL to verify the change, step-by-step actions, test credentials if login is required, and the expected result - without this, the reviewer has to guess what to verify and review takes longer. After setup, every push to main triggers a notification; the review reads the spec and changed files, fixes issues directly in code, and opens a PR, typically within 24 hours.

Use it when wanting human review of AI-generated code pushed to GitHub, when a project spec exists that reviewers should check implementation against, when review feedback should arrive as a follow-up PR with suggested fixes rather than just comments, and when granting the Vibers service collaborator access to the repository is acceptable. What it checks: spec compliance, security against the OWASP Top 10, AI hallucinations such as fake APIs or imports, logic bugs, and UI issues. What it does not check: code style (use ESLint or Prettier instead), performance benchmarks, or full QA (use Playwright or Cypress instead) - it is not a replacement for those. It requires a GitHub repository and the collaborator grant, the referenced spec must stay accessible to the review workflow, and turnaround depends on the external service rather than being instant. A non-GitHub path exists too: send code and spec directly via Telegram.

Input is the pushed commit's changed files plus the linked spec document. The GitHub Action config:

name: Vibers Code Review
on:
  push:
    branches: [main]

jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 2
      - uses: marsiandeployer/vibers-action@v1
        with:
          spec_url: 'https://docs.google.com/document/d/YOUR_SPEC_ID/edit'
          telegram_contact: '@your_telegram'

Output is a follow-up pull request containing the reviewer's fixes, plus a Telegram notification when it's ready. Pricing is per-review, no subscription: a $1/hour promo tier (full review plus PRs with fixes, in exchange for honest feedback) and a $15/hour standard tier (full review plus a security audit and priority turnaround).

Integrates via a GitHub Action (marsiandeployer/vibers-action@v1) triggered on push, a GitHub collaborator grant for the marsiandeployer account, a publicly accessible spec document, and Telegram for review-ready notifications; supports JS/TS, Python, React, Next.js, Django, Flask, and other languages or frameworks as long as the project is on GitHub. Feedback can be sent directly:

curl -X POST https://vibers.onout.org/feedback \
  -H 'Content-Type: application/json' \
  -d '{"message": "Your question or issue", "repo": "https://github.com/you/your-repo"}'

Teams shipping AI-generated code who want an external human reviewer checking spec compliance, security, and hallucinated APIs before merging, delivered as a ready-to-merge PR rather than just review comments.