Karpathy Guidelines

Karpathy Guidelines

Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations on LLM coding pitfalls. For deeper background on the jagged intelligence that causes these failure modes, see the [[ghosts-vs-animals]] concept in the LLM wiki.

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

The Three Failure Modes (Why These Principles Exist)

Karpathy identified three recurring problems when LLMs generate code autonomously. See references/karpathy-agent-failure-modes.md for the full Karpathy quotes.

The root cause: LLMs exhibit jagged intelligence — genius polymath and confused grade-schooler simultaneously. The principles below catch the 10-year-old mistakes while letting the PhD-level reasoning run.

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. Prefer Native Tools Over External Infrastructure

Don't build scripts, databases, or repos when the tools you already have can do the job.

A common failure pattern: a skill or workflow depends on infrastructure that doesn't exist — a SQLite database, a Python script directory, a self-evolution repo, a cron job. Before adding any external dependency:

• Does an existing tool already cover this? (session_search, skill_manage, memory, delegate_task, web_search)
• Could the workflow run inside the conversation instead of as a standalone script?
• Is the external dependency actually installed/set up on this system, or would you be writing code for an imaginary runtime?

The test: If removing the external dependency would collapse the approach, the approach was wrong. A Karpathy-style solution uses only tools that exist. It does not require setup steps, installation, or infrastructure provisioning.

Examples: • Instead of "query state.db for past failures" → use session_search • Instead of "run a Python script to classify errors" → use the agent's own reasoning • Instead of "clone a self-evolution repo" → use skill_manage(patch) directly • Instead of "write metrics to a JSON file" → use mnemosyne_remember

4. 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.

5. 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.