Migration guide

Most teams already have someform of QA — Cursor Bugbot for code review, Chromatic / Percy for visual regression, Devin for autonomous coding, or just a hand-written script that takes screenshots. This page maps each tool to the equivalent CodeLoop surface and gives you a step-by-step migration path.

From Cursor Bugbot

Bugbot reviews PRs and posts comments. CodeLoop runs before the PR exists (inside the agent loop) andon the PR (via the GitHub Action) — same evidence, two surfaces, with a concrete confidence number instead of natural-language opinion.

Mapping

Migration steps

1. Run npx codeloop initin your project — this adds the IDE rule files and the MCP config.
2. Drop in the GitHub Action workflow from codeloop-verify@v1.
3. Disable Bugbot on the repo (or leave it — they don't conflict). The PR comment from CodeLoop replaces the value Bugbot was providing.

From Devin

Devin is an autonomous coder that runs full sessions on its own. CodeLoop is what makes those sessions trustworthy. They are complementary — Devin (or any other agent) writes the code, CodeLoop verifies that it works.

Mapping

Wire CodeLoop into Devin by giving it the same MCP config as Cursor / Claude Code — Devin's session bootstrapper supports MCP servers natively.

From Chromatic / Percy

Chromatic and Percy are visual regression services. CodeLoop covers the same use case (visual review) plus design compare, plus interaction recordings, plus the rest of the verify loop.

Mapping

Migration steps

1. Export your Chromatic / Percy baselines (right-click » save PNG, or use their export API).
2. Drop them under .codeloop/baselines/<screen>/<viewport>.png and commit.
3. Update your CI to call npx codeloop verify instead of chromatic / percy exec.
4. Move ignore regions from your Chromatic config to visual_review.ignore_regions in .codeloop/config.json.

From manual QA

If you currently rely on a human running through the app before each release, CodeLoop encodes that checklist as a machine-readable spec.

Migration steps

1. Take your existing checklist (the one in Notion / Confluence / someone's head) and write each line as a testable claim in docs/E2E_TEST_CHECKLIST.md.
2. Wire each claim to a verify input — usually a Playwright / Maestro test or a screenshot capture.
3. Run codeloop_gate_check with acceptance_path: docs/E2E_TEST_CHECKLIST.md. Each line becomes an item in the acceptance_criteria_met gate.
4. The first run will reveal missing coverage — CodeLoop tells you exactly which checklist items have no evidence yet.

From hand-rolled screenshot scripts

If you already have a Playwright / Cypress / Puppeteer script that takes screenshots and diffs them with pixelmatch, you can keep it — wire it in as a CodeLoop plugin:

{
  "plugins": [
    {
      "name": "screenshot_suite",
      "command": "node",
      "args": ["scripts/screenshot.js"],
      "detect_file": "scripts/screenshot.js",
      "parse_output": "exit_code"
    }
  ]
}

The plugin's output now feeds the gate score and surfaces in the dashboard. Once you trust the loop, replace the script with the built-in codeloop_capture_screenshot calls.

Side-by-side: when does each shine?

Coexistence

CodeLoop is intentionally additive. You can keep Bugbot for code-style review, keep Chromatic for the hosted snapshot UI, and add CodeLoop for the verify loop without removing either. The only place to be careful is the GitHub Action sticky comment — if both Bugbot and CodeLoop post on the same PR, give them different sticky-comment labels (set sticky-id: codeloopin the action input) so they don't fight for the same comment.

Related

• Side-by-side comparison page
• Quick Start
• Core concepts