Closed-Loop Verification 🔄

BrowserTrack features an automated closed-loop verification engine for both runtime bug fixes and visual layout fixes.

🐞 Bug Fix Verification (verify_incident) #

When an agent fixes a code bug, it triggers verify_incident:

1. Browser Reload: Reloads the active browser tab (or navigates to the incident route).
2. Error Monitoring: Watches for the incident fingerprint during an observation window (e.g. 2000ms).
3. Probe Evaluation: Evaluates custom verification probes:
   • no_incident: Confirms zero runtime errors occurred.
   • element_exists: Checks if a specific selector exists in the DOM.
   • element_visible: Confirms the target is rendered and visible.
   • text_contains: Asserts expected text content.
   • route_is: Verifies current SPA route.
4. After Screenshot: Automatically captures an after-fix screenshot.
5. Verdict: Returns VERIFIED, FAILED, or INCONCLUSIVE.

// Example MCP call
await client.callTool("verify_incident", {
  incidentId: "inc_19a31a28",
  expect: [
    { type: "no_incident" },
    { type: "element_visible", selector: "#user-card" }
  ]
});

📐 Layout & Visual Note Verification (verify_note) #

When an agent modifies CSS or HTML to resolve visual feedback, it triggers verify_note:

1. Target Element Inspection: Queries element visibility, bounding rect, and computed styles.
2. Viewport Overflow Probe: Checks if rect.right > viewportWidth or if container has unintended horizontal scroll.
3. Geometry Diff: Calculates before vs after dimensional delta (widthDiff, heightDiff, overflowFixed).
4. After Screenshot: Captures an updated visual snapshot.
5. Verdict & Auto-Resolution: If all checks pass and overflow is eliminated, returns VERIFIED.

// Example MCP call
await client.callTool("verify_note", {
  noteId: "note_30a89599",
  observationWindowMs: 1000
});