Agent Memory

Agent memory lets Specwright get faster and smarter over time. After exploring a page once, the planner remembers selectors so re-runs use verification mode (5 browser calls instead of 20).

Memory locations

.claude/agent-memory/
├── playwright-test-planner/
│   └── MEMORY.md        ← Discovered selectors per URL
└── playwright-test-healer/
    └── MEMORY.md        ← Fix patterns for recurring failures

Planner memory

After each successful exploration, the planner writes a ## Key Selectors section:

## Key Selectors: @LoginPage (http://localhost:3000/login)
Last explored: 2026-04-20

| Element         | Locator                              |
|-----------------|--------------------------------------|
| Email input     | page.getByTestId('login-email')      |
| Password input  | page.getByTestId('login-password')   |
| Sign In button  | page.getByTestId('login-submit')     |
| Error message   | page.getByTestId('login-error')      |

On the next run for the same URL, the planner enters verification mode — navigates, takes one snapshot, confirms each selector still resolves, re-discovers only selectors that have drifted.

Healer memory

After fixing a test failure, the healer records the pattern so similar failures in the future are resolved immediately.

When to clear memory

Memory is never cleared automatically. Clear it manually when:

• You've done a major redesign and selectors have changed completely
• Memory from a stale URL is causing verification mode to skip real exploration

# Clear planner memory for a fresh exploration
echo "# Playwright Test Planner — Agent Memory" > \
  .claude/agent-memory/playwright-test-planner/MEMORY.md