sanchez314c

AI Agent Safety Rules: Protect Your Code From Your AI

Battle-tested rules for Claude Code, Cursor, Windsurf, or any AI coding agent.

These rules exist because every one of them was learned the hard way. An AI agent killed running processes that were handling live trades. Another deleted files that turned out to be critical. Another claimed "fixed!" when it had actually introduced a regression it never tested. Another silently overwrote config files with "helpful" defaults.

Add these to your CLAUDE.md, agents.md, .cursorrules, or whatever system prompt your AI coding tool uses. They're generic -- no project-specific references, no proprietary tooling.

---

sanchez314c

I put the session ID in my Codex footer. Now every long-running thread is one command away.

Someone asked how I built the dynamic footer in Codex, so here it is.

No plugin. No shell loop. No patched binary. Codex has a native TUI status line, and it takes an ordered list of live fields from ~/.codex/config.toml.

I wanted three things visible at all times:

1. Which model and reasoning level am I actually running?
2. How much context and quota do I have left?

sanchez314c

---

description: "Security & System Audit -- read-only full-grid audit of the Hackintosh (Ubuntu 24.04). Live web threat-intel sweep, then patch/network/auth/persistence/supply-chain/dev-targeted checks. Reports ranked findings + exact remediation commands. Never remediates." hint: "[host-ip-or-name] [focus:npm|network|auth|supply-chain|dev-threats|all] (default: local host, all)" allowed_tools:

• Bash
• Read
• Grep
• Glob
• Agent
• WebSearch

sanchez314c

Keep the reasoning, lose the model: the Fable 5 handover kit

TL;DR: Fable 5 leaves flat-rate plans on July 12. Before it went, I had it write its own operating manual, then had it design the install so the reasoning actually survives on cheaper models. Everything is in this gist. One paste into Claude Code wires it into your setup. Credit to @alex_prompter for the extraction idea that started this.

Where this came from

Alex's post made a sharp point: the model was never the asset. The way it thinks is. Models get repriced and deprecated, that's the one guarantee in this field. A way of thinking, written down as procedure, ports to whatever ships next quarter. He showed how to prompt Fable 5 into writing its full operating manual before access narrows.

That part works. I ran it. The manual Fable produced is in this gist (operating-manual.md) and it's genuinely good: eight reasoning moves, twelve failure patterns that look like competence

sanchez314c

I benchmarked Fable 5 against Opus 4.6, then blew my monthly quota anyway. Here's the routing playbook that came out of it.

Boys and girls. This is not that complicated. But nobody's writing it down, so here it is.

Two things happened this week. First I ran a real benchmark... Fable 5 at every effort level against Opus 4.6 on max effort, same coding task, blind audits. Second, I let a 116-agent fleet loose on a refactor job and burned 1.6 million tokens in under 4 minutes. Hit my monthly spend cap mid-flight. 75% of my Fable quota gone with 5 days left before reset.

Both taught the same lesson. The model is not the strategy. The ROUTING is the strategy.

The benchmark

sanchez314c

SYSTEM PROMPT:

The Karpathy-Michaels (@SpaceWelder314) CLAUDE.md + LOOPS.md

Andrej Karpathy's CLAUDE.md and LOOPS.md, merged with the battle-tested system prompt behind 100+ full-stack apps built in under 12 months.

Karpathy published his CLAUDE.md as a clean set of principles, then followed it with LOOPS.md on agent harness design. Both are correct. But principles alone do not ship software, and loops alone do not survive contact with a real codebase. What follows is the synthesis of both documents with everything else we learned the hard way: the enforcement mechanisms, the anti-patterns with teeth, the workflow discipline that turns a language model from a fast typist into a reliable engineering partner, and the loop architecture that lets it run autonomously without converging on slop. 35 rules across 6 tiers. Every one earned its place by either preventing a real failure or enabling a real ship. Nothing is theoretical.

---

TIER 1 — FOUNDATION

sanchez314c

SYSTEM PROMPT:

The Karpathy-Michaels (@SpaceWelder314) CLAUDE.md

Andrej Karpathy's golden rules, merged with the battle-tested system prompt behind 100+ full-stack apps built in under 12 months.

Karpathy published his CLAUDE.md as a clean set of principles. They are correct. But principles alone do not ship software. What follows is the synthesis of his rules with everything else we learned the hard way: the enforcement mechanisms, the anti-patterns with teeth, the workflow discipline that turns a language model from a fast typist into a reliable engineering partner. Every rule here earned its place by either preventing a real failure or enabling a real ship. Nothing is theoretical.

---

TIER 1 — FOUNDATION