Verify Your App with stably verify

stably verify checks whether your application works correctly by describing what it should do in plain English. It launches an AI agent that navigates your app with a real browser, interacts with it, takes screenshots, and reports a structured PASS / FAIL / INCONCLUSIVE verdict. No test files are generated — it verifies and reports.

stably verify "the login form accepts email and password and redirects to /dashboard"

To run verification in a Stably-hosted browser, add --browser=cloud or set STABLY_CLOUD_BROWSER=1. See Cloud Browsers for Stably CLI.

How It Works

1. You run: stably verify "users can add items to cart and see updated total"
   |
2. Interactive preflight (auto-detects your dev server and credentials)
   |
3. AI agent launches a browser and systematically verifies each requirement
   - Navigates to the app
   - Interacts with the UI (clicks, fills forms, scrolls)
   - Takes screenshots as evidence at each step
   - Checks DOM state, network responses, and visual output
   |
4. Agent reports a structured verdict: PASS, FAIL, or INCONCLUSIVE
   |
5. CLI prints a formatted summary and sets the exit code
   - Exit 0 = PASS
   - Exit 1 = FAIL
   - Exit 2 = INCONCLUSIVE

Prerequisites

Stably CLI

Install and initialize the Stably CLI in your project.

Usage

# Verify a feature works
stably verify "the login form accepts email and password and redirects to /dashboard"
stably verify "the login form accepts email and password and redirects to /dashboard" --browser=cloud

# Verify with a specific starting URL
stably verify "the pricing page shows 3 tiers" --url https://localhost:3000/pricing

# Set a budget cap (default: $5)
stably verify "checkout flow completes successfully" --max-budget 10

Options

Option	Description
`-u, --url <url>`	Starting URL to verify against (auto-detected from localhost if omitted)
`--max-budget <dollars>`	Maximum budget in USD for the session (default: `5`)
`--no-interactive`	Skip interactive preflight (for CI or agent use)

Exit Codes

Code	Verdict	Meaning
`0`	PASS	All requirements in the prompt were verified
`1`	FAIL	One or more requirements were not met
`2`	INCONCLUSIVE	Could not determine pass or fail (app unreachable, auth wall, etc.)

Example Output

PASS

VERIFICATION PASSED

Prompt: "users can add items to cart and see updated total"

Steps performed:
1. Navigated to /products
2. Clicked "Add to Cart" on first product -> cart badge updated to 1
3. Clicked "Add to Cart" on second product -> cart badge updated to 2
4. Opened cart -> shows 2 items with correct total of $45.98

Evidence:
- Screenshot: products page with items visible
- Screenshot: cart showing 2 items and $45.98 total
- DOM check: .cart-total element contains "$45.98"

FAIL

VERIFICATION FAILED

Prompt: "users can add items to cart and see updated total"

Steps performed:
1. Navigated to /products
2. Clicked "Add to Cart" on first product -> cart badge updated to 1
3. Clicked "Add to Cart" on second product -> cart badge stayed at 1

Failure reason:
Cart badge did not update after adding second item. Expected "2", saw "1".
The cart total also only reflects one item ($22.99) instead of two.

Evidence:
- Screenshot: cart badge showing "1" after adding 2 items

Using stably verify with AI Agents

stably verify is especially powerful when used inside AI coding agents. The agent writes code, then runs stably verify to check if the feature actually works — creating a build-then-verify iteration loop.

Install the Skill

Install the stably-verify skill to teach your agent how to use the command:

npx skills add https://github.com/stablyai/agent-skills --skill stably-verify

Agent Iteration Loop

Agent makes code changes
        ↓
  stably verify "feature description"
        ↓
  ┌─ PASS?  → Done
  └─ FAIL?  → Agent reads the verdict, fixes code, re-verifies

Example: Claude Code

You:    Build a search bar that filters products by name

Agent:  I'll add the search component to the products page...
        [writes code]
        Let me verify this works.

        $ stably verify "the products page has a search bar that filters
          products by name as you type" --url http://localhost:3000/products

        VERIFICATION FAILED
        Steps performed:
        1. Navigated to /products — 12 product cards visible
        2. Found search input and typed "blue"
        3. Product list did not change — still showing all 12 products

        Failure reason: Search input exists but typing does not filter.

Agent:  The search input isn't wired up to filter state. Let me fix that...
        [fixes code]

        $ stably verify "the products page has a search bar that filters
          products by name as you type" --url http://localhost:3000/products

        VERIFICATION PASSED
        Steps performed:
        1. Navigated to /products — 12 product cards visible
        2. Typed "blue" in search bar → 2 products shown
        3. Cleared search → all 12 products restored

Agent:  Search bar is working correctly.

Example: Cursor / Windsurf

Ask your agent naturally — the skill is triggered automatically:

"Use stably verify to check that the checkout flow works end to end"
"Verify the dashboard shows the correct charts after my changes"

The Skill Prompt

stably-verify SKILL.md

---
name: stably-verify
description: |
  Verify that an application works correctly using stably verify. Use when
  an AI agent has made code changes and needs to validate the feature works
  in a real browser. Triggers on: "verify this works", "stably verify",
  "check if this works", "validate my changes", "verify my feature".
license: MIT
metadata:
  author: stably
  version: '1.0.0'
---

# Verify App Behavior with stably verify

Use `stably verify` to check that your application behaves correctly against
a natural language description. The command launches an AI agent that navigates
your app in a real browser, interacts with it, and reports a PASS/FAIL/INCONCLUSIVE
verdict. No test files are generated.

## Pre-flight

Always run `stably --version` first. If not found, install with
`npm install -g stably` or use `npx stably`.

## Usage

stably verify "description of expected behavior"
stably verify "pricing page shows 3 tiers" --url http://localhost:3000/pricing
stably verify "checkout completes" --max-budget 10
stably verify "login works" --no-interactive

## Options

--url <url>            Starting URL (auto-detected if omitted)
--max-budget <dollars> Budget cap in USD (default: $5)
--no-interactive       Skip preflight prompts (for CI/agents)

## Exit Codes

0 = PASS (all requirements verified)
1 = FAIL (one or more requirements not met)
2 = INCONCLUSIVE (unable to determine — app not reachable, auth wall, etc.)

## Iteration Loop

After making code changes, verify with stably verify. If the verdict is FAIL:

1. Read the failure reason and evidence from the output
2. Fix the application code based on the specific failure
3. Re-run stably verify with the same prompt
4. Repeat until PASS

Important: stably verify does not create or modify files. It only observes
and reports. Fix the code yourself based on its findings.

## Composing in Scripts

stably verify "login works" && echo "Feature verified!"

stably verify "checkout completes" ; code=$?
if [ $code -eq 0 ]; then echo "PASS";
elif [ $code -eq 1 ]; then echo "FAIL";
else echo "INCONCLUSIVE"; fi

Combining with Other Stably Commands

Command	Purpose	When to Use
`stably verify`	Verify app behavior against a description	Quick feature validation, no test files needed
`stably create`	Generate Playwright test files	Building lasting regression tests
`stably test`	Run existing Playwright tests	Running your committed test suite
`stably fix`	Auto-fix failing tests	Tests broke and need AI-powered repair

A typical workflow with an AI agent:

Build and verify

Use stably verify to quickly check that your feature works as you build it. No test files to manage — just describe what should happen.

Create lasting tests

Once the feature is stable, use stably create to generate Playwright test files for regression coverage.

Maintain with test + fix

For future changes, run stably test to check the suite and stably fix to auto-repair failures.

CI/CD Integration

stably verify works in CI pipelines with --no-interactive:

# .github/workflows/verify.yml
- name: Verify feature
  run: npx stably verify "checkout flow completes successfully" --no-interactive
  env:
    STABLY_API_KEY: ${{ secrets.STABLY_API_KEY }}
    STABLY_PROJECT_ID: ${{ secrets.STABLY_PROJECT_ID }}

Use exit codes to gate deployments:

stably verify "all critical flows work" --no-interactive && deploy.sh

Next Steps

CLI Quickstart

Get set up with the Stably CLI

CLI Commands

Full command reference

Create Tests

Generate Playwright test files with AI

Auto-fix Tests

AI-powered test fixing

​How It Works

​Prerequisites

Stably CLI

​Usage

​Options

​Exit Codes

​Example Output

​PASS

​FAIL

​Using stably verify with AI Agents

​Install the Skill

​Agent Iteration Loop

​Example: Claude Code

​Example: Cursor / Windsurf

​The Skill Prompt

​Combining with Other Stably Commands

​CI/CD Integration

​Next Steps

CLI Quickstart

CLI Commands

Create Tests

Auto-fix Tests

How It Works

Prerequisites

Usage

Options

Exit Codes

Example Output

PASS

FAIL

Using stably verify with AI Agents

Install the Skill

Agent Iteration Loop

Example: Claude Code

Example: Cursor / Windsurf

The Skill Prompt

Combining with Other Stably Commands

CI/CD Integration

Next Steps