Submitting Solution

The file contains SWEA_BASE_URL, SWEA_ID, and SWEA_PW. Use SWEA_BASE_URL as {base_url} throughout this skill (all URLs below use this prefix). Never hardcode credentials or URLs.

Step 0: Ensure Playwright MCP is connected

Playwright MCP is normally kept disabled to save memory. Before using this skill, ensure it is connected.

1. Attempt to call any Playwright MCP tool (e.g., browser_snapshot).
2. If the tool call succeeds, Playwright is already connected — proceed to Step 1.
3. If the tool call fails (tool not found / MCP server not connected): a. Inform the user: "The Playwright MCP server is not connected. Please enable it in your agent's MCP settings and restart, then re-run this skill." b. Stop here — do not proceed until the user restarts and re-invokes the skill.

Headless mode: The Playwright MCP server is configured to run with --headless. Do not remove this flag.

How to use Playwright MCP

This skill uses the Playwright MCP server (mcp__playwright__* tools) to control a headless browser. Key tools and their usage:

General workflow:

1. browser_navigate to a URL.
2. browser_snapshot to see the page content and element ref numbers.
3. browser_click on a ref to interact with an element (input field, button, link).
4. browser_type to enter text into the focused element.
5. Repeat snapshot → click/type as needed.

Step 1: Read the solution code

Read the solution file at ./problem_bank/$0/python/solution_$1.py and store its contents. This will be pasted into the code editor.

Step 2: Log in

If {base_url} is https://swexpertacademy.samsung.com (company SSO):

• Skip login entirely — SSO handles authentication automatically.
• browser_navigate to {base_url} and browser_snapshot to verify the page loads successfully.

Otherwise (default ID/PW login):

1. browser_navigate to {base_url}/main/identity/anonymous/loginPage.do.
2. browser_snapshot to find the ID input field, password input field, and login button.
3. browser_click on the ID input ref, then browser_type with the SWEA_ID value.
4. browser_click on the password input ref, then browser_type with the SWEA_PW value.
5. browser_click on the login button ref.
6. browser_wait_for for 2 seconds, then browser_snapshot to confirm login succeeded (URL should no longer contain loginPage).

Step 3: Navigate to the problem

1. browser_navigate to {base_url}/main/code/problem/problemList.do.
2. browser_snapshot to find the search input (placeholder: "문제 번호, 키워드").
3. browser_click on the search input ref, then browser_type with the problem ID $0 and submit: true.
4. browser_wait_for for 2 seconds, then browser_snapshot to see search results.
5. Find the link for problem $0 and browser_click on it.
6. browser_wait_for for 2 seconds to let the problem detail page load.

Step 4: Open the solving page

1. browser_snapshot to find the "문제 풀기" (or "문제 풀러 가기" on company mirror) link on the problem detail page (upper-right area, calls javascript:goProblem()).
2. browser_click on it. This opens a new tab at /main/solvingProblem/solvingProblem.do.
3. browser_tabs with action: "select" and index: 1 to switch to the new solving tab.
4. browser_wait_for for 2 seconds to let the solving page fully load.

Step 5: Select Python language

1. browser_snapshot to find the Language combobox.
2. Check if "Python 3.7 (PyPy 7.3.9)" is already selected.
3. If not, browser_select_option on the Language combobox to select the Python option.

Step 6: Paste the solution code

The code editor is a CodeMirror instance. The underlying <textarea> is hidden behind CodeMirror's overlay and cannot be clicked directly. Use browser_run_code to set the code via the CodeMirror API:

async (page) => {
  const code = `<solution_code_here>`;
  await page.evaluate((codeStr) => {
    const cm = document.querySelector('.CodeMirror').CodeMirror;
    cm.setValue(codeStr);
  }, code);
}

Important: Escape backticks and ${ sequences in the solution code before embedding it in the JavaScript template literal. Alternatively, pass the code as a variable to page.evaluate().

After setting the code, browser_snapshot to verify the editor shows the correct code (check the first and last lines).

Step 7: Submit

1. browser_snapshot to find the "제출" (Submit) link at the bottom right.
2. browser_click on the "제출" link.
3. A confirmation dialog appears: "제출 가능 횟수가 1회 감소합니다. 정말로 제출하시겠습니까?"
4. browser_snapshot to find the "확인" (OK) button.
5. browser_click on "확인" to confirm submission.

Step 8: Read the verdict

After confirming, the judge runs and a result dialog appears with the verdict. browser_snapshot to read the dialog text.

Parse the verdict from the dialog:

The dialog may also contain details like "N개의 테스트케이스 중 M개가 맞았습니다" (M out of N test cases passed). Extract this information if available.

Click "확인" to dismiss the result dialog.

Step 9: Clean up

1. Close the solving tab: browser_tabs with action: "close".
2. Call browser_close to shut down the browser.

Result

Report the verdict back to the orchestrator in this format:

Submit — {verdict} (details: {M}/{N} test cases passed, if available)

The orchestrator (solving-problem) will use this verdict to decide the next step (congratulate on Pass, or proceed to Reflect on failure).