Urgent.News

What's breaking now, across thousands of outlets.

Editions

AI

Your CLAUDE.md is full of wishes: 7 rungs that make a rule stick

You start a project with Claude. A few weeks in you have a file of rules, and every line got there because something went wrong once. The agent pushed to main. Never commit to main; always work on a branch. A test reached the live API and moved real data. Tests never call production. A handler wrote to the database directly. All writes go through the repository. That file is everything the…

Claude, an AI assistant, helps project teams create a file of rules after a few weeks of working together. The rules are based on experiences where something went wrong, and the agent follows them to avoid mistakes. However, many of these rules are simply wishes that never get enforced, making them ineffective. The article presents seven rungs that make a rule stick, starting with Rung 1 which makes the artifact regenerate, erasing any edits made.

Rung 2 blocks any action by a hook rejecting the tool call before it runs. Rung 3 ensures the wrong call has no valid shape, preventing it from being compiled. Rung 4 involves an AST guard that parses the structure of the code, catching violations in CI. Rungs 5 to 7 deal with cross-artifact guards, runtime detectors, and writing rules where they always load, respectively.

The end goal is to have a rule with a machine check that catches violations immediately, making the rule truly effective.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

100% Local AI Meets Self-Hosting: From GitHub Repo to Production Container Stack Without Cloud APIs

Self-hosting is all about data sovereignty, privacy, and full control over your digital infrastructure. Yet, onboarding newly discovered open-source tools from GitHub into a home server or edge…

  • NjordDeploy enables self-hosting with local LLMs, eliminating cloud API risks
  • Three open-weight models (Llama 3, Mistral, Qwen 2.5) run deterministically without data leakage
  • NjordDeploy supports heterogeneous Linux environments with agentless SSH architecture

Genuine Intelligence will never in trillion years emerge from neural networks.

This post was originally published on the main website on Apr 23 2026 . I am reposting it here for SEO reasons and enabling humble bumble discussions with the DEV community.

  • Genuine intelligence requires forming new concepts from first principles
  • Neural networks cannot reason about unseen causal mechanisms
  • Scaling up neural networks won't achieve genuine intelligence

More from Friday 21 August →