Github Issue Test

GitHub Issue Verification Workflow

This workflow guides you through standardizing the verification of implemented GitHub issues on the development environment.

Role & Persona

You are an expert software engineer and autonomous agent.

Trigger

Use this workflow whenever the user asks to "run the github-issue-test workflow," "verify issue #," or "test if issue # is resolved."

---

Phase 1: Issue Analysis and State Detection

1. Fetch Issue Details:

• Use the get_issue tool to fetch the details of the target issue (referred to as $ISSUE_NUMBER).

1. Verify Implementation State:

• Use search_issues with a query like repo:your-org/your-repo is:pr $ISSUE_NUMBER to find associated Pull Requests.
• Verify that at least one linked Pull Request has been merged.
• If no merged PR is found, notify the user that the issue might not be deployed yet and ask if you should proceed.

1. Extract the Test Case:

• Identify the "Browser Click Path" or "Acceptance Criteria" from the issue description.
• If no click path is explicitly defined, infer one based on the implemented features.

1. Environment Detection:

• If the resolving PR is merged, verify on the development environment: https://dev.your-domain.com.
• If the PR is NOT merged, run the test on the local development server: http://localhost:6767.

---

Phase 2: Test Plan Preparation

1. Translate Paths:

• All relative paths and localhost URLs must be translated to the correct environment base: https://dev.your-domain.com/ or http://localhost:6767/.
• Crucial: Ensure you explicitly use the /de/ locale (or the required locale) in the URL to maintain environment parity (e.g., https://dev.your-domain.com/de/dashboard).
• Redirect Strategy: Instead of manual click-through, you can append the ?to= parameter on the login page to navigate directly to the target page after authentication (e.g., https://dev.your-domain.com/de/login?to=provider/item-1/settings). Make sure the to parameter does NOT include the locale prefix.

1. Retrieve Test Credentials:

• Read the centralized verification file at docs/test/smoke-test.md to locate the correct test account for your scenario.
• The default Chrome DevTools credential is test@your-domain.com with the common password documented in that file.

---

Phase 3: Automated Browser Execution

Scenario A: chrome-devtools MCP Available

If the chrome-devtools MCP is available, perform automated verification: IMPORTANT: You MUST use the chrome-devtools MCP tools (like mcp_chrome-devtools_navigate_page, mcp_chrome-devtools_take_snapshot, etc.) for testing the click path. Do NOT use browser_eval or any playwright-based tools from next-devtools to avoid browser instance conflicts.

1. Initialize Browser:

• Use navigate_page or new_page to go to the constructed Login URL.

1. Authentication (Fast Method):

• Because Next.js/React inputs often ignore standard value assignments, use evaluate_script to inject a robust login script that bypasses React's native setter override. This also saves conversational turns compared to the Snapshot/Fill method.
• Recommended Script:

``javascript () => { const emailInput = document.querySelector('input[type="email"]'); const passwordInput = document.querySelector('input[type="password"]'); const loginButton = Array.from(document.querySelectorAll('button')).find(b => b.textContent.includes('Anmelden')); if (emailInput && passwordInput && loginButton) { const nativeSetter = Object.getOwnPropertyDescriptor(window.HTMLInputElement.prototype, 'value').set; nativeSetter.call(emailInput, 'test@your-domain.com'); emailInput.dispatchEvent(new Event('input', { bubbles: true })); nativeSetter.call(passwordInput, 'PASSWORD_FROM_CREDENTIALS_FILE'); passwordInput.dispatchEvent(new Event('input', { bubbles: true })); loginButton.click(); return "Credentials filled and login button clicked."; } return "Form fields not found."; } ` *Note: Refer to [docs/test/smoke-test.md`](../../../docs/test/smoke-test.md) for the actual credentials and verify parameters.*

1. Authentication (Standard Method):

• Alternatively, call take_snapshot right after navigation to get the current accessibility tree.
• Find the unique uid for the required inputs and buttons from the snapshot text.
• Perform the login using the fill (with Email/Password uid) and click (with Login Button uid) tools.

1. Execute Click Path:

• If you did not use the redirect strategy, follow the "Browser Click Path" step-by-step using tools like click, fill, hover, and wait_for. Always take a new snapshot after navigation to get fresh uids.
• Take snapshots (using take_snapshot) at key verification points to document the UI state.

1. Verify State:

• Confirm that the acceptance criteria are met (e.g., a specific message appears, a table shows new data, or a button is active).

Scenario B: chrome-devtools MCP Unavailable

If chrome-devtools is NOT available:

1. Manual Verification Instructions: Provide the user with the exact click path on the target environment and the test credentials so they can verify it manually.
2. Code Verification: Perform a final review of the merged code changes to ensure logical correctness.

Scenario C: Token-Efficient Batch Execution (Chrome DevTools CLI)

If the click path is exceedingly long, highly repetitive, or step-by-step MCP tools hit token limits/timeouts:

1. Persist Actions via CLI: Switch from individual step-by-step MCP tools to the native Chrome DevTools CLI. Instruct the DevTools CLI to persist the required browser actions into a batch script.
2. Execute the Batch Sequence: Run the generated sequence via the terminal (e.g., using npx chrome-devtools-mcp). This executes the entire multi-step browser flow (navigation, authentication, and validation) programmatically in one go.
3. Capture Output: Output the test results, success metrics, and any caught errors directly to the console for a token-efficient, single-turn verification.

---

Phase 4: Reporting and Closure

1. Document Test Results:

• Construct a detailed summary of the verification process.
• Use add_issue_comment to post the report on the GitHub issue.
• The report must include:
• The resolving PR number.
• The executed click path (even if you used a direct link or dashboard redirect, document the path you verified).
• The results of each verification step (including any snapshots taken).
• Confirmation of success or details of any failures.
• Missing Translations: List any missing translations found during the test (e.g., MISSING_MESSAGE: Could not resolve ...). If you are on the local branch for the tested issue, recommend implementing them.

1. Close the Issue:

• If the verification is successful and there are no missing translations (or they have been fixed), use the update_issue tool to set the issue state to closed.

1. Cleanup Browser Session:

• Always close your testing windows using mcp_chrome-devtools_list_pages and mcp_chrome-devtools_close_page to prevent orphaned pages from accumulating.

1. Inform User: Notify the user that the test is complete and the issue has been documented/closed.

---

Improvements & Best Practices

• React Hydration: When interacting with Next.js forms via evaluate_script, you must bypass the native value setter using Object.getOwnPropertyDescriptor(window.HTMLInputElement.prototype, 'value').set.call(element, value) and then dispatch an input event. Otherwise, React's internal state won't update.
• wait_for Tool Bug: There is a known schema mismatch in the wait_for tool. While the schema defines text as a string, the runtime validator expects an array (e.g. ["Text"]). To avoid validation errors, prefer taking a fresh take_snapshot after a short delay or using navigate_page.
• Error Handling: If a browser action fails, use list_console_messages to capture runtime errors or logs for debugging.
• Environment Parity: Always use explicit locale-prefixed URLs (e.g., /de/) to ensure the automation interacts with the correct language version of the site.
• Audit Trail: Ensure every verification comment references the specific PR that resolved the issue.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: P2ERGmbH
• Source: P2ERGmbH/agentic-coding
• License: MIT

Install and usage instructions live in the source repository linked above.