test-live

name	test-live
description	Interactive live testing of a running web application via Playwright MCP browser automation
user_invocable	true
usage	/workflow:test-live <url> [--user=<email>] [--pass=<password>] [--instructions=<text>]
arguments	[{"name":"url","required":true,"description":"URL of the running web application to test (e.g., http://localhost:8080)"},{"name":"user","required":false,"description":"Login username/email for authenticated testing"},{"name":"pass","required":false,"description":"Login password for authenticated testing"},{"name":"instructions","required":false,"description":"Specific testing instructions or scenarios to verify"}]

Live E2E Testing

Test a running web application interactively using Playwright MCP browser automation. This is a lightweight, standalone testing command - no workflow state, no gates, just direct browser testing with structured reporting.

Usage

/workflow:test-live http://localhost:8080
/workflow:test-live http://localhost:3000                                    # authenticated runs read login from ./.creds
/workflow:test-live https://staging.myapp.com --instructions="Test the checkout flow with a $50 item"

Input

$ARGUMENTS

---

Execution Steps

Step 1: Parse Input

Extract from arguments:

• URL (required): First argument or value after the command
• --user / --pass: optional override credentials — prefer the gitignored ./.creds file (see Credentials below) over passing secrets on the command line
• --instructions: Specific test scenarios (optional, rest of text after flags)

If URL is missing, ask the user for it.

Credentials. For authenticated testing, read the login from the project's ./.creds file first (gitignored — test creds live there; never commit it or echo its contents into logs/commits/PRs). Only fall back to --user/--pass, or ask the user, if ./.creds is missing or lacks the needed account. Never print the password or pass it inline where it would be logged.

Step 2: Validate URL

URL scheme validation (MANDATORY): Only http:// and https:// URLs are accepted. Reject file://, ftp://, javascript:, data:, and any other scheme.

URL_SCHEME=$(echo "<url>" | grep -oP '^[a-z]+(?=://)')
if [ "$URL_SCHEME" != "http" ] && [ "$URL_SCHEME" != "https" ]; then
  echo "ERROR: Only http:// and https:// URLs are supported"
fi

Check reachability:

curl -sS -o /dev/null -w "%{http_code}" --max-time 10 <url>

• If unreachable: warn user but allow proceeding (app may need auth)
• If reachable: proceed

Step 3: Check Playwright MCP Availability

Try using a Playwright MCP tool (e.g., browser_navigate to the URL).

If Playwright MCP tools are NOT available:

1. Check if .claude/.mcp.json exists in the project root
2. If it exists, read it and check for a playwright entry
3. If no playwright config, add it:

# Read existing config or create new
existing = Read(".claude/.mcp.json") or "{}"
config = json.loads(existing)

if "mcpServers" not in config:
    config["mcpServers"] = {}

if "playwright" not in config["mcpServers"]:
    config["mcpServers"]["playwright"] = {
        "command": "npx",
        "args": ["@playwright/mcp@latest", "--headless"]
    }
    Write(".claude/.mcp.json", json.dumps(config, indent=2))

4. Also check user-level MCP config at ~/.claude/.mcp.json as an alternative location
5. After writing the config, inform the user:

Playwright MCP server has been configured in .claude/.mcp.json.

To activate it, please restart your Claude Code session, then run:
  /workflow:test-live <url> [options]

Alternatively, you can add the Playwright MCP server manually:
  claude mcp add playwright -- npx @playwright/mcp@latest --headless

If Playwright MCP tools ARE available: proceed to Step 4.

Step 4: Spawn Web Tester Agent

Build the prompt for the web-tester agent based on parsed input:

prompt_parts = [
    f"## Task\nTest the web application at {url}\n",
]

if auth_required:  # ./.creds present, or --user/--pass override provided
    prompt_parts.append("""## Authentication
Read the login credentials from the project's ./.creds file (do NOT print them),
authenticate first, then proceed with testing. If ./.creds is absent and override
credentials were provided, use those — without echoing the password.
""")

if instructions:
    prompt_parts.append(f"""## Test Instructions
Follow these specific testing instructions:
{instructions}

After completing the instructed tests, report findings.
""")
else:
    prompt_parts.append("""## Test Scope
No specific instructions provided. Perform exploratory testing:
1. Test all visible pages from the navigation
2. Test all forms (valid and invalid input)
3. Test interactive elements (buttons, modals, dropdowns)
4. Check for console errors on each page
5. Verify images and resources load correctly
6. Test basic user flows (if authenticated)
""")

Agent(
    subagent_type="workflow:web-tester",
    model="sonnet",
    prompt="\n".join(prompt_parts)
)

Step 5: Report Results

After the web-tester agent completes, present the results to the user:

• Summary of pages and forms tested
• Issues found with severity
• Recommendations for fixes
• Any screenshots taken

If the agent found critical issues, highlight them prominently.