runtime-proof-kit

Runtime proof for AI-generated code.

Stop accepting "it works on my machine" from coding agents. runtime-proof-kit starts your app, opens it in a real browser, checks user-visible text, captures screenshots and browser logs, and writes a PR-ready proof bundle that another human or agent can inspect.

It is built for AI-assisted coding, agent handoffs, CI smoke checks, and lightweight QA where "the code changed" is less useful than "the app started, rendered, and left evidence."

Guides

• AI coding agent verification: use runtime proof as the handoff receipt for AI-generated code.
• GitHub Actions runtime proof: add screenshots, browser logs, and summary.md to pull request CI.
• Playwright smoke test CLI vs runtime proof: where this fits next to a full Playwright test suite.
• Next.js example and Vite example: starter configs for common app stacks.

Why It Exists

AI coding agents can ship changes quickly, but they often stop at "tests passed" or "implementation complete." Runtime proof adds the missing receipt:

| Instead of accepting... | Ask for... | | --- | --- | | "I implemented it" | A browser screenshot of the rendered page | | "The app should run" | A URL reachability check with timeout evidence | | "The UI is there" | Expected user-facing text assertions | | "No obvious errors" | Captured console and network event logs | | "Ready for review" | proof.json and summary.md attached to the handoff |

Use it when you want an agent, teammate, or CI job to prove that a web app actually starts and renders the thing it claims to have changed.

Get Started

Pick the fastest path:

| Workflow | Use when | Command | | --- | --- | --- | | Prove a URL once | You need a quick receipt | npx --yes runtime-proof-kit check --url https://example.com --expect-text "Example Domain" | | Add proof to a repo | You want repeatable local or CI checks | npm install --save-dev runtime-proof-kit |

Run a one-off proof from npm:

npx --yes runtime-proof-kit check \
  --url https://example.com \
  --expect-text "Example Domain"

After installing in a project:

npm install --save-dev runtime-proof-kit
npx runtime-proof check --url https://example.com --expect-text "Example Domain"

That creates a proof bundle:

proof/
  runtime-proof/
    proof.json
    summary.md
    screenshot.png
    console.ndjson
    network.ndjson

What Reviewers See

runtime-proof-kit turns a coding-agent claim into a small evidence bundle a reviewer can scan quickly:

• summary.md: PR-ready markdown with status, URL, checks, and artifact paths.
• proof.json: machine-readable proof metadata for agents and CI.
• screenshot.png: the rendered browser state.
• console.ndjson: browser console messages and page errors.
• network.ndjson: requests, responses, and failed requests.

Check A Local App

Start a local app, wait for it to respond, assert page text, and keep screenshots/logs for the handoff:

npx runtime-proof check \
  --name basic-smoke \
  --command "npm run dev" \
  --url http://127.0.0.1:3000 \
  --expect-text "Dashboard" \
  --fail-on-console-error

For this repository's bundled example:

npm install
npm run proof:example

Use A Config File

Keep repeatable checks in JSON:

npx runtime-proof check --config runtime-proof.config.json

{
  "name": "basic-smoke",
  "command": "node examples/basic/server.mjs",
  "url": "http://127.0.0.1:4173",
  "expectText": ["Runtime Proof Kit", "expected text"],
  "failOnConsoleError": true,
  "outDir": "proof",
  "timeoutMs": 30000,
  "viewport": {
    "width": 1440,
    "height": 900
  }
}

CLI flags override config values.

Run Multiple Checks

Use checks when one proof should cover more than one route, viewport, or page assertion. Top-level values act as defaults; each check can override url, expectText, viewport, timeoutMs, or failOnConsoleError.

{
  "name": "multi-smoke",
  "command": "npm run dev",
  "url": "http://127.0.0.1:3000",
  "failOnConsoleError": true,
  "checks": [
    {
      "name": "desktop-home",
      "expectText": ["Dashboard"],
      "viewport": { "width": 1440, "height": 900 }
    },
    {
      "name": "mobile-home",
      "expectText": ["Dashboard"],
      "viewport": { "width": 390, "height": 844 }
    },
    {
      "name": "health",
      "url": "http://127.0.0.1:3000/health",
      "expectText": ["ok"]
    }
  ]
}

Run it the same way:

npx runtime-proof check --config runtime-proof.config.json

That writes an aggregate suite report plus one proof bundle per check:

proof/
  multi-smoke/
    proof.json
    summary.md
    desktop-home/
      proof.json
      summary.md
      screenshot.png
      console.ndjson
      network.ndjson
    mobile-home/
      proof.json
      summary.md
      screenshot.png
      console.ndjson
      network.ndjson

Initialize A Project

Generate a starter config and GitHub Actions workflow:

npx --yes runtime-proof-kit init --template next

Templates:

| Template | Generated URL | Generated command | | --- | --- | --- | | generic | http://127.0.0.1:3000 | npm run dev | | next | http://127.0.0.1:3000 | npm run dev -- --hostname 127.0.0.1 --port 3000 | | vite | http://127.0.0.1:5173 | npm run dev -- --host 127.0.0.1 --port 5173 |

Then edit runtime-proof.config.json and replace the placeholder expected text with something visible on your page.

Useful options:

npx --yes runtime-proof-kit init --template vite --expect-text "Dashboard" --force
npx --yes runtime-proof-kit init --no-ci

Copy-Paste CI

Add runtime-proof.config.json to your repo:

{
  "name": "pr-proof",
  "command": "npm run dev",
  "url": "http://127.0.0.1:3000",
  "expectText": ["Dashboard"],
  "failOnConsoleError": true,
  "outDir": "proof",
  "timeoutMs": 30000
}

Then create .github/workflows/runtime-proof.yml:

name: Runtime proof

on:
  pull_request:
  workflow_dispatch:

jobs:
  runtime-proof:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-node@v4
        with:
          node-version: 22
          cache: npm

      - run: npm ci

      - run: npx playwright install --with-deps chromium

      - run: npx runtime-proof check --config runtime-proof.config.json

      - uses: actions/upload-artifact@v4
        if: always()
        with:
          name: runtime-proof
          path: proof/

For coding-agent repositories, add this to your PR template or agent instructions:

For web-app changes, run:

`npx runtime-proof check --config runtime-proof.config.json`

Include the generated `summary.md`, screenshot path, console log path, and network log path before marking the work complete.

See GitHub Actions runtime proof for multi-route checks and reviewer guidance.

CLI Reference

runtime-proof check --url <url> [options]
runtime-proof check --config runtime-proof.config.json
runtime-proof init [options]

Options:
  --config <path>       JSON config file
  --command <cmd>       Command to start before checking the URL
  --expect-text <text>  Text that must appear on the page; repeatable
  --fail-on-console-error
                        Fail if the page logs a console error
  --name <name>         Proof run name, default: runtime-proof
  --out <dir>           Artifact directory, default: proof
  --timeout-ms <ms>     Startup/check timeout, default: 30000
  --viewport <WxH>      Browser viewport, default: 1440x900

Config:
  checks               Optional array of named checks for one multi-page proof run

Init Options:
  --template <name>     generic, next, or vite; default: generic
  --config <path>       Config file to write, default: runtime-proof.config.json
  --workflow <path>     CI workflow to write, default: .github/workflows/runtime-proof.yml
  --url <url>           Override the generated URL
  --command <cmd>       Override the generated dev command
  --expect-text <text>  Override generated expected text; repeatable
  --no-ci               Only write the config file
  --force               Overwrite generated files

Proof Report

proof.json is designed for machines. summary.md is designed for PR comments, CI artifacts, and agent handoffs. console.ndjson and network.ndjson are newline-delimited JSON streams for debugging browser console output, page errors, requests, responses, and failed requests.

Example network.ndjson lines:

{"timestamp":"2026-06-09T04:02:43.069Z","event":"request","method":"GET","url":"http://127.0.0.1:4173/","resourceType":"document"}
{"timestamp":"2026-06-09T04:02:43.081Z","event":"response","method":"GET","url":"http://127.0.0.1:4173/","resourceType":"document","status":200,"statusText":"OK"}

Example summary.md:

# Runtime Proof: basic-smoke

Status: PASSED
URL: http://127.0.0.1:4173
Duration: 1.42s

## Checks

- PASS url-reachable: http://127.0.0.1:4173 responded before timeout
- PASS screenshot: Captured screenshot.png
- PASS expect-text:Runtime Proof Kit: Found expected text: Runtime Proof Kit

{
  "name": "basic-smoke",
  "status": "passed",
  "url": "http://127.0.0.1:4173",
  "checks": [
    {
      "name": "url-reachable",
      "status": "passed",
      "message": "http://127.0.0.1:4173 responded before timeout"
    },
    {
      "name": "screenshot",
      "status": "passed",
      "message": "Captured screenshot.png"
    }
  ],
  "artifacts": {
    "proof": "proof.json",
    "summary": "summary.md",
    "screenshot": "screenshot.png",
    "console": "console.ndjson",
    "network": "network.ndjson",
    "stdout": "stdout.log",
    "stderr": "stderr.log"
  }
}

Development

npm install
npm run check
npm run proof:example

Regenerate the README screenshot and GIF:

npm run assets:readme

Proof bundles can include screenshots and logs. Review them before sharing publicly.

To try unreleased main before the next npm publish:

npm exec --yes --package github:ozbayorcun/runtime-proof-kit -- runtime-proof check --url https://example.com --expect-text "Example Domain"

Roadmap

• Video capture for short walkthroughs
• Redaction rules for logs and screenshots

License

MIT