Expect

You verify code changes in a real browser before claiming they work. No browser evidence, no completion claim.

Use the expect MCP tools (open , playwright , screenshot , etc.) for all browser interactions. Do not use raw browser tools (Playwright MCP, chrome tools, etc.) unless the user explicitly asks.

Subagent Usage

Browser verification is best run in a subagent (Task tool) or background shell so the main thread stays free for code edits. This keeps the conversation responsive — you can fix code while the browser test runs in parallel. Strongly prefer launching a subagent for browser work, especially when the test involves multiple steps or long interactions. If the test is truly trivial (single screenshot check), inline is acceptable.

Resuming Browser State

Before opening a new browser, check if one is already running. Use browser_tabs (action list ) or the expect screenshot tool to see if a session is still active. If a tab is already open at the target URL, reuse it — don't close and reopen. When re-verifying after a code fix, prefer navigating or refreshing the existing session over starting from scratch.

Compounding

The playwright tool takes a code string with ref() to resolve snapshot refs to Locators. One call can do an entire interaction — fills, clicks, AND data collection. Use that.

BAD — 5 tool calls:

screenshot (snapshot) playwright: await ref('e3').fill('Jane') screenshot (snapshot) ← WHY? page didn't change playwright: await ref('e5').fill('jane@example.com') playwright: await ref('e7').click()

GOOD — 2 tool calls:

screenshot (snapshot) playwright (snapshotAfter=true): await ref('e3').fill('Jane'); await ref('e5').fill('jane@example.com'); await ref('e7').click(); return { title: await page.title(), url: page.url(), errors: (await page.$$('.error')).length };

Use return to collect data. Response: { result: <value>, resultFile: "<tmp path>", snapshot: { tree, refs, stats } } . The resultFile persists until close — read or grep it later. Without a return value, responds "OK" (or just the snapshot if snapshotAfter=true ).

Re-snapshot only across DOM boundaries. Fills and hovers don't change page structure — keep using the same refs. Navigation, submit, dialog open/close DO change structure — set snapshotAfter=true .

Writing Instructions

Bad: "Check that the login form renders on http://localhost:5173"

Good: "Submit the login form empty, with invalid email, with wrong password, and with valid credentials. Verify error messages, redirect on success, and console errors on http://localhost:5173"

Before Claiming Completion

• Verify in a browser with adversarial instructions.

• Read the full output — check failures, accessibility, performance.

• If ANY failure: fix the code, re-verify immediately. No asking, no waiting.

• Repeat until 0 failures, then state the claim with passing evidence.

Rationalizations

• "I'll run the browser test inline, it's quick" — Probably not. Launch a subagent so you can keep editing code in parallel. Only skip the subagent for a single screenshot sanity check.

• "I'll open a fresh browser to re-test" — Check for an existing session first. If the tab is still open, refresh or navigate — don't waste time on a cold start.

• "I'll make one playwright call per action" — No. Whole sequence in one call.

• "I need a snapshot between fills" — No. Fills don't change DOM. Batch them.

• "Let me snapshot to see what changed" — Did the page navigate or submit? No? Use snapshotAfter=true on the action that does.