diff --git a/AGENTS.md b/AGENTS.md
index e7859738..cc721728 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -12,17 +12,87 @@
 ## Code Quality
 - Keep cyclomatic complexity under 10 per function
 - Aim for files under 300 lines
+- All files must follow single responsibility principle (one clear purpose per module)
+- Extract complex logic into reusable utilities/services
 
 ## Testing
+
+### Standard Testing (Ongoing)
 - Write unit tests for all utility functions
 - Minimum 80% coverage on new code
 - Use Vitest for unit tests
 
+### AI-Friendly Refactoring Testing Strategy (v1.10.16+)
+**Scope:** Full test coverage before refactoring to prevent UI/functionality regression.
+
+**Backend Testing (Pytest)**
+- Target: 85%+ coverage (unit + integration tests)
+- Structure: `backend/tests/` with suites for routers, models, auth, AI pipeline
+- Coverage tools: `pytest backend/tests/ --cov=backend --cov-report=html`
+- Test types: Unit (functions), Integration (full API workflows), Fixtures (mocked auth, in-memory DB)
+
+**Frontend Testing (Vitest)**
+- Target: 80%+ coverage (components, hooks, snapshots)
+- Structure: `frontend/tests/` with component tests, hook tests, integration workflows
+- Coverage tools: `npm test -- --coverage`
+- Test types: Component rendering, Hook state/side effects, Snapshot tests for UI layouts
+
+**E2E Testing (Playwright)**
+- Target: Critical user workflows automated
+- Workflows: Login, Scan → Match → Stock adjustment, New Item (AI), Admin config, Offline sync
+- Runtime: ~30 min total
+- Command: `npx playwright test`
+
+**Test-First Approach**
+- All tests written and passing BEFORE refactoring any code
+- Tests serve as gating condition: no code changes if tests fail
+- Functional preservation: Zero behavior changes post-refactor
+- Regression prevention: Manual checklist + automated tests catch UI breakage
+
 ## Security
 - Store all secrets in inventory.env — never hardcode credentials
 - Never log API keys, tokens or passwords
 - Validate all user input before processing
 
 ## Git Conventions
-- Use conventional commits (feat:, fix:, docs:, etc.)
+- Use conventional commits (feat:, fix:, docs:, refactor:, test:, etc.)
 - Keep PRs under 400 lines of diff when possible
+- Refactoring commits: `refactor: split {module} into smaller modules`
+- Testing commits: `test: add {suite} coverage for {module}`
+- All commits must have passing test suite before merge
+
+## Refactoring Guidelines (AI-Friendly Modularity)
+
+### Refactoring Principles
+- **Target:** Break monolithic files into focused modules (<300 lines each)
+- **Priority Order:** Backend routers → Components → Pages
+- **No Behavior Changes:** Every refactor must pass 100% of existing tests
+- **Gating Rule:** No refactor commit unless all tests pass (pre + post)
+- **Regression Prevention:** Manual checklist + automated tests validate UI integrity
+
+### Refactoring Process
+1. Ensure full test coverage exists (backend 85%, frontend 80%)
+2. Run complete test suite (all tests PASS)
+3. Refactor module: split into smaller files/services
+4. Run test suite again (all tests PASS)
+5. Manual browser validation: UI unchanged, all buttons/features work
+6. Commit with test results in message body
+7. Move to next module
+
+### Module Extraction Patterns
+- **Backend Routers:** Split into router (endpoints only) + service (business logic) + validators (input validation)
+- **Components:** Split into orchestrator component + sub-components + custom hooks for state/logic
+- **Pages:** Extract page logic into custom hooks, move forms into separate components
+- **Utilities:** Consolidate repeated logic into `lib/` or `services/` modules
+
+### Validation Checklist (Post-Refactor)
+- [ ] All pytest tests passing (backend coverage 85%+)
+- [ ] All vitest tests passing (frontend coverage 80%+)
+- [ ] All e2e workflows passing (Playwright)
+- [ ] Login works (LDAP + local)
+- [ ] Scanner functional (scan → match → stock adjustment)
+- [ ] AI extraction working (new item → AI popup → validation)
+- [ ] Admin page responsive and functional
+- [ ] No console errors in browser
+- [ ] Mobile responsive (320px, 768px, 1024px+ viewports)
+- [ ] Keyboard navigation works (focus indicators visible)