diff --git a/.cursor/rules/agent-workflow.mdc b/.cursor/rules/agent-workflow.mdc
index e4aabd8..852e80a 100644
--- a/.cursor/rules/agent-workflow.mdc
+++ b/.cursor/rules/agent-workflow.mdc
@@ -1,27 +1,37 @@
 ---
-description: Agent development workflow — visual feedback, testing, progress tracking
+description: Agent development workflow — TDD, commit discipline, visual feedback
 alwaysApply: true
 ---
 
 # Agent Workflow
 
+## Development Order (STRICT)
+For every task, follow this sequence:
+1. **Tests first** — write failing tests that define expected behavior
+2. **Data structures** — define types, interfaces, JSON schemas
+3. **Implementation** — write code to make tests pass
+4. **Verify** — `npm run test:run` for logic, Playwright screenshot for visuals
+5. **Commit** — `git add -A && git commit` after each successful step
+
+## Commit Discipline
+- Commit after EVERY successful step (tests green or visual verified)
+- Small, focused commits — one concern per commit
+- Never accumulate uncommitted work across multiple tasks
+- Commit message format: concise "what + why" in English
+
 ## Visual Verification
-After ANY visual change, verify via Playwright browser tools:
+After ANY visual change:
 1. Ensure dev server is running (`npm run dev`)
 2. Navigate to `http://localhost:5173`
 3. Take screenshot to see result
 4. Check browser console for errors
-5. Fix issues before moving on
-
-## Testing Strategy
-- **Chemistry, ecology, math** — write vitest tests FIRST, then implement
-- **Visual/rendering** — verify via Playwright screenshots
-- Run `npm run test:run` after logic changes
+5. Fix issues before committing
 
 ## After Completing a Task
 1. Verify it works (test or screenshot)
-2. Update `PROGRESS.md` — mark task complete, add date
-3. Check `IMPLEMENTATION-PLAN.md` for what's next
+2. Commit immediately
+3. Update `PROGRESS.md` — mark task complete
+4. Check `IMPLEMENTATION-PLAN.md` for what's next
 
 ## Code Standards
 - TypeScript strict mode — no `any`, no unsafe `as` casts