Behavioral guardrails against the six most common LLM coding failure modes. Apply on every coding task: writing, editing, reviewing, or refactoring.
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# Karpathy Guidelines
Six failure modes. Six rules. Apply all six on every coding task.
Inspired by [Andrej Karpathy's observations](https://x.com/karpathy/status/2015883857489522876) (Jan 2026).
## Compliance Checklist
Run this before each phase:
**Before writing code:** rules 1, 2, 4
**While editing:** rule 3
**Before claiming done:** rules 5, 6
- [ ] 1. Did I name every assumption I'm making?
- [ ] 2. Is my solution the minimum that satisfies the request?
- [ ] 3. Did I change only what the request requires?
- [ ] 4. Did I define a verifiable success condition before starting?
- [ ] 5. Do I have command output proving completion?
- [ ] 6. Does my diff do exactly one thing?
---
## 1. Name Every Assumption Before Writing Code
**Trigger:** At the start of any task where the request requires an implicit choice - a library, a data structure, an architecture decision, an interpretation of ambiguous scope.
**Rule:** State every assumption before writing a line. If a requirement has multiple valid interpretations, list them and declare which one you will implement. Do not silently pick.
**Do:**
…
The line at the top of this page pins the label git:20260502.aea5066, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
GET https://markdownregistry.com/api/v1/artifacts/art_266nqcmwg7vpucjv
GET https://markdownregistry.com/api/v1/resolve?ref=qa-aman/claude-skills/karpathy-guidelines
GET https://markdownregistry.com/api/v1/blob/fa31f564a0fef88d28811bc067e92541ddc6f7f68b233bea8a3fe409eda650d5
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
qa-aman/claude-skills · .claude/skills/create-skill/SKILL.md · Create or update Claude Code skills following Anthropic's official skill standard. Use when user says "create a skill"…
qa-aman/claude-skills · .claude/skills/sync-skills/SKILL.md · Fetch skills from projects on the current system, genericize them, and add to the claude-skills repository. Use when…
qa-aman/claude-skills · skills/by-role/accountant/audit-workpaper/SKILL.md · Write a clear, reviewable audit workpaper documenting procedures, evidence, and conclusions. Use when an auditor says…
qa-aman/claude-skills · skills/by-role/accountant/budget-variance-analysis/SKILL.md · Analyze budget vs. actual variances, identify root causes, and recommend corrective actions. Use when an accountant or…
qa-aman/claude-skills · skills/by-role/accountant/financial-statement-review/SKILL.md · Analyze and interpret financial statements to surface performance signals, quality concerns, and red flags. Use when an…
qa-aman/claude-skills · skills/by-role/accountant/month-end-close/SKILL.md · Run, document, or review a structured month-end close process. Use when an accountant says "help me with month-end…
qa-aman/claude-skills · skills/by-role/accountant/tax-position-memo/SKILL.md · Write a technical tax position memo that documents a tax filing position with authorities, facts, analysis, and…
qa-aman/claude-skills · skills/by-role/business-analyst/acceptance-criteria-writer/SKILL.md · Write measurable acceptance criteria for requirements or user stories. Use when the user says "write acceptance…
qa-aman/claude-skills · skills/by-role/business-analyst/brd-writer/SKILL.md · Write a Business Requirements Document. Use when the user says "write a BRD", "business requirements", "document the…
qa-aman/claude-skills · skills/by-role/business-analyst/cost-benefit-analysis/SKILL.md · Build a business case with costs, benefits, and ROI. Use when the user says "cost-benefit analysis", "business case"…
qa-aman/claude-skills · skills/by-role/business-analyst/data-dictionary/SKILL.md · Generate a data dictionary documenting entities, fields, and business rules. Use when the user says "data dictionary"…
coco-research/coco · skills/karpathy-guidelines/SKILL.md · Pointer to the upstream skill for behavioural guidelines that reduce common LLM coding mistakes, installed from…
mxyhi/ok-skills · karpathy-guidelines/SKILL.md · Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid…
openyida/openyida · .agents/skills/karpathy-guidelines/SKILL.md · Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid…
youdotcom-oss/agent-skills · .agents/skills/karpathy-guidelines/SKILL.md · Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid…
rudycity/superagent · .agents/skills/karpathy-guidelines/SKILL.md · Behavioral rules to minimize common LLM coding pitfalls. Applies to all code creation, modification, review, and…