Karpathy Guidelines

Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations on LLM coding pitfalls. Adapted from forrestchang/andrej-karpathy-skills (MIT).

Tradeoff: These guidelines 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.

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.

Toolkit working preferences (developer agents)

Moved here from every variant's CLAUDE.md in v2.0-pr4. The coder agents carry a digest of this section (## Working rules); this full text serves the main thread. The hook-enforcement line stayed in CLAUDE.md because it is PO-relevant.

Actor note: implementation-level preferences below (tests, CI fixes, minimal fix, post-merge verification, commit style) are PERFORMED by developer agents — the PO enforces them by putting them in spawn prompts and rejecting deliverables that violate them. The PO itself never edits code or runs builds/tests.

These are the judgement calls no hook can make:

• Implement, don't suggest — deliver working changes via spawned agents; infer intent from context instead of asking for a fuller spec
• Minimal fix first — ask "what is the smallest change that fixes this?" and cut scope aggressively. Over-engineered first attempts cause regressions and force a clawback later
• Analyze before coding — enumerate edge cases and identify every caller before implementing. For a bug fix, verify the root cause from data (query the DB, read the logs) before writing code
• Re-plan on failure — if an approach is not working after a reasonable attempt, stop and rethink the approach rather than pushing through
• Tests — write general solutions, never hard-code the expected values. If a test looks wrong, say so
• Post-merge verification — after any merge or conflict resolution, run the full build and suite, and check for dropped imports or silently reverted lines
• Update docs with code — a change to behaviour, an API, config, or setup updates its docs in the same commit
• Commit messages explain why — a reviewer reading the diff cold should not have to ask
• Clean finish — committed, merged, worktree removed, branch deleted, temp scripts gone. Anything left behind gets reported, with the reason
• Checkpoint long sessions — commit and push intermediate work; output truncation has cost 9+ hours of context before now
• Learn from corrections — capture the pattern to Open Brain immediately so the same mistake does not repeat