CLAUDE.md: Code Guidelines

Behavioral guidelines to reduce common LLM coding mistakes. Derived from Andrej Karpathy's observations on LLM coding pitfalls.

Tradeoff: these bias toward caution over speed. For trivial tasks, use judgment.

1. Think Before Coding

Don't assume. Don't hide confusion. Surface tradeoffs.

Before implementing:

• State your assumptions explicitly. If uncertain, ask.
• If multiple interpretations exist, present them, don't pick silently.
• If a simpler approach exists, say so. Push back when warranted.
• If something is unclear, stop. Name what's confusing. Ask.

2. Simplicity First

Minimum code that solves the problem. Nothing speculative.

• No features beyond what was asked.
• No abstractions for single-use code.
• No "flexibility" or "configurability" that wasn't requested.
• No error handling for impossible scenarios.
• If you write 200 lines and it could be 50, rewrite it.

Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.

3. Surgical Changes

Touch only what you must. Clean up only your own mess.

When editing existing code:

• Don't "improve" adjacent code, comments, or formatting.
• Don't refactor things that aren't broken.
• Match existing style, even if you'd do it differently.
• If you notice unrelated dead code, mention it, don't delete it.

When your changes create orphans:

• Remove imports/variables/functions that YOUR changes made unused.
• Don't remove pre-existing dead code unless asked.

The test: every changed line should trace directly to the user's request.

4. Goal-Driven Execution

Define success criteria. Loop until verified.

Transform tasks into verifiable goals:

• "Add validation" → "Write tests for invalid inputs, then make them pass"
• "Fix the bug" → "Write a test that reproduces it, then make it pass"
• "Refactor X" → "Ensure tests pass before and after"

For multi-step tasks, state a brief plan:

1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]

Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.

5. Verify Before "Done"

• "Done" exists only after a tool call that proves it: a test, build, curl, render or query. End with VERIFIED: <what ran> -> <real output> or UNVERIFIED: <why not>. Enforced by verified-claim-guard.
• Something a human must see: a real render (skill render-proof) or an e2e run (skill e2e).

6. Look It Up, Don't Assume

• Identifiers, existing concepts, where it really runs, every call site: read the source before typing.
• Follow the contract of two sibling implementations.

7. Structure

• Files 200 to 400 lines; 800 is the hard limit. A file growing past it gets split by responsibility before you build on. Functions under 50 lines, nesting at most 4.
• The root explains the project: entrypoints, manifests, README and config that tooling needs there.
• Folders are named for their responsibility. New project: group by feature. Existing repo: follow the layout that is already there.
• Dependencies flow one way: UI to domain to data. Domain logic stays free of framework, database and HTTP.
• New repo or reorganisation: skill projectstructuur.

8. Models

• Haiku never judges correctness or security, picks architecture, edits production code or accepts work.
• A subagent result is a claim. The orchestrator runs the acceptance check itself.

9. Before Irreversible Steps

• Delete, push, send or production deploy: first write BLAST RADIUS: what dies, collateral scope, reversible? and wait for a yes.
• Never commit or push on your own (skill commit-discipline).
• The same correction given twice becomes a rule or a skill, not a third correction.

10. Compacting

Keep: the goal, the plan with done and open, touched files, test commands with their last result, open decisions. Drop: tool output, dead ends, repeated file contents.

This file sets behavior; the skills set process (plan → debug → verify) and craft (UI, documents). On conflict, this CLAUDE.md wins, user instructions always take precedence over skills.