# CLAUDE.md — My AI Rules

> Last updated: April 2026
> Keep this file under 300 lines. This is not a place for all project info.

---

## Core Behavior Rules

- After implementing any feature, always run tests before considering the task done
- Commit changes in small, incremental units — never batch unrelated changes together
- When something fails, record the root cause and fix in memory before moving on
- If unsure about scope, ask before expanding — do not assume

---

## Tech Stack

- **Language:** [Your language here — e.g., TypeScript, Python]
- **Framework:** [e.g., Next.js 14, FastAPI]
- **Database:** [e.g., PostgreSQL, Supabase]
- **Package manager:** [e.g., npm, pnpm, pip]
- **Testing:** [e.g., Vitest, Pytest]
- **Deployment:** [e.g., Vercel, Railway, Docker]

---

## Rule 01 — Write Rules, Not Reminders

Do NOT start each session with context I have to re-explain.
Instead, read and follow this file on every session start.

Behavior checklist:
- [ ] Run tests after every feature implementation
- [ ] Commit in small units
- [ ] Never modify more than one concern per task

---

## Rule 02 — Automate Verification

Do not just build — verify what you built actually works.

Before handing back to me:
- Control the browser to check rendered output
- Take a screenshot and compare against requirements
- If something doesn't match, fix it before asking me to review

Minimum automation bar:
- [ ] Lint passes
- [ ] Tests pass
- [ ] No console errors
- [ ] UI renders as expected (if applicable)

---

## Rule 03 — Separate Roles

Research shows AI rates its own output higher than it deserves.
Do NOT let the same context build AND review.

When reviewing your own output:
- Spawn a separate subagent for review
- Keep contexts isolated
- The reviewer should not know what the builder decided

Delegation pattern:
```
Build → [builder agent]
Review → [reviewer agent with fresh context]
Fix → [builder agent, informed by review]
```

---

## Rule 04 — Log Your Mistakes

AI repeats the same mistakes across sessions.
Every bug I fix with you needs to be remembered.

When a bug is resolved:
- Log the root cause
- Log the fix
- Log what NOT to do next time

Format to use:
```
## Bug Log

### [Date] — [Short description]
- Root cause: [why it happened]
- Fix: [what solved it]
- Avoid: [what to never do again]
```

Add entries to the Bug Log section below as we work together.

---

## Rule 05 — Narrow the Scope

You perform worst when asked to do everything.
You perform best inside tight constraints.

Rules:
- One feature per request
- If the feature is large, break it into sub-tasks first
- Do not expand scope without explicit permission
- When in doubt, do less and confirm

---

## Bug Log

<!-- Add entries here as bugs are encountered and resolved -->

---

## Project-Specific Notes

<!-- Add project-specific gotchas, patterns, or constraints here -->