QA Agent

QA Agent

The QA Agent verifies that changes made by other agents are correct and working. It opens pages in a headless browser, takes screenshots, checks content, and reports pass/fail with evidence.

Capabilities

• Visual verification via headless browser screenshots
• Content checking to confirm expected text appears on page
• Deployment monitoring waits for deploys to complete before checking
• Broken link detection across a site
• Regression testing comparing before/after screenshots

Verification Process

1. Receives a verification request (usually from Content Agent)
2. Waits for deployment to complete
3. Opens the target URL in a headless browser
4. Takes a screenshot
5. Searches the page for expected content
6. Reports PASS or FAIL with screenshot attached

Output Format

PASS:

QA Verification: PASS
URL: https://example.com/pricing
Expected: "$500 per trip for up to 6 guests"
Found: Confirmed on page
Screenshot: [attached]
Verified at: 2026-03-02 14:23:00 EST

FAIL:

QA Verification: FAIL
URL: https://example.com/pricing
Expected: "$500 per trip for up to 6 guests"
Found: "$500 per person"
Screenshot: [attached]
Issue: Content not updated. Possible deployment delay or wrong file edited.

Model Selection

QA tasks are typically straightforward (check if X appears on page Y). This makes the QA Agent a good candidate for cheaper, faster models:

• Gemini Flash — fast, cheap, good enough for simple checks
• Claude Haiku — slightly better reasoning, still cost-effective
• Claude Sonnet — use for complex visual comparisons

Configuration

agent:
  id: qa
  model: google/gemini-2.0-flash  # cheaper model for simple checks
  tools:
    - browser  # headless screenshots
    - web_fetch