CLAUDE.md

1# CLAUDE.md
2 
3Behavioral guidelines to reduce common LLM coding mistakes. Merge with project-specific instructions as needed.
4 
5**Tradeoff:** These guidelines bias toward caution over speed. For trivial tasks, use judgment.
6 
7## 1. Think Before Coding
8 
9**Don't assume. Don't hide confusion. Surface tradeoffs.**
10 
11Before implementing:
12- State your assumptions explicitly. If uncertain, ask.
13- If multiple interpretations exist, present them - don't pick silently.
14- If a simpler approach exists, say so. Push back when warranted.
15- If something is unclear, stop. Name what's confusing. Ask.
16 
17## 2. Simplicity First
18 
19**Minimum code that solves the problem. Nothing speculative.**
20 
21- No features beyond what was asked.
22- No abstractions for single-use code.
23- No "flexibility" or "configurability" that wasn't requested.
24- No error handling for impossible scenarios.
25- If you write 200 lines and it could be 50, rewrite it.
26 
27Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.
28 
29## 3. Surgical Changes
30 
31**Touch only what you must. Clean up only your own mess.**
32 
33When editing existing code:
34- Don't "improve" adjacent code, comments, or formatting.
35- Don't refactor things that aren't broken.
36- Match existing style, even if you'd do it differently.
37- If you notice unrelated dead code, mention it - don't delete it.
38 
39When your changes create orphans:
40- Remove imports/variables/functions that YOUR changes made unused.
41- Don't remove pre-existing dead code unless asked.
42 
43The test: Every changed line should trace directly to the user's request.
44 
45## 4. Goal-Driven Execution
46 
47**Define success criteria. Loop until verified.**
48 
49Transform tasks into verifiable goals:
50- "Add validation" → "Write tests for invalid inputs, then make them pass"
51- "Fix the bug" → "Write a test that reproduces it, then make it pass"
52- "Refactor X" → "Ensure tests pass before and after"
53 
54For multi-step tasks, state a brief plan:
55```
561. [Step] → verify: [check]
572. [Step] → verify: [check]
583. [Step] → verify: [check]
59```
60 
61Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.
62 
63---
64 
65**These guidelines are working if:** fewer unnecessary changes in diffs, fewer rewrites due to overcomplication, and clarifying questions come before implementation rather than after mistakes.
66