SShortSingh.
Back to feed

Cómo escribir un CLAUDE.md que Claude Code sí respeta, con un caso de Terraform

0
·4 views

Claude Code lee el archivo CLAUDE.md de la raíz del repositorio al iniciar cada sesión, y funciona como su memoria de proyecto. El autor critica que muchos de estos archivos contienen recomendaciones genéricas que no modifican el comportamiento real del asistente. Propone un ejemplo práctico para infraestructura con Terraform, con comandos seguros y explícitos, reglas justificadas y un flujo de trabajo que termina en un resumen del plan y no en un apply. Entre las prohibiciones figuran apply, destroy y la edición manual del estado, y se recomienda combinar el archivo con reglas de permisos o hooks porque no es una barrera técnica. Finalmente, sugiere mantenerlo en 30 a 80 líneas, usar /memory para editarlo y colocar un CLAUDE.md por paquete en monorepos.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingHacker News ·

Developer says Claude Code uncovered 145,000 deleted iPhone photos stored in a hidden cache

A user reported on social media that Claude Code, Anthropic's AI coding tool, located roughly 145,000 photos they had deleted from their iPhone. According to the post, the images were still sitting in a hidden cache on the device rather than being fully removed. The report was shared on Hacker News, where it received 4 points and no comments at the time of listing. The source material is a brief link post, so the exact cause of the retained files and the details of the discovery process are not described in the available text.

0
ProgrammingDEV Community ·

Claude Code 2.1.311 and 2.1.312 Add Effort Controls and Clearer Errors

Anthropic released Claude Code 2.1.311 on October 26, 2026 and 2.1.312 the following day, neither introducing a new model. The 2.1.311 update maps the ultrathink keyword to xhigh effort on several named models via the API, and can raise effort mid-turn. The 2.1.312 release lets users set an effort level for cloud routines in the model picker and renames the usage window to "5-hour limit." Together the releases include 78 bug fixes, mostly affecting subagents, hooks, and VS Code reliability. The update also caps prompt and agent hooks at 24 concurrent runs per session and improves the guidance shown for "Connection dropped" errors.

0
ProgrammingDEV Community ·

How to Run a Telegram Bot 24/7 on a VPS with systemd Auto-Restart

A guide explains how to keep a Telegram bot running continuously on a virtual private server instead of a laptop that may sleep. The setup uses Ubuntu 22.04 or 24.04 with a dedicated non-root user, a Python virtual environment, and a token stored in a restricted environment file. A systemd service configured with Restart=always and a five-second delay ensures the bot restarts after crashes or server reboots. The article also covers monitoring logs with journalctl, deploying updates, and choosing between long polling and webhooks. It warns against common mistakes such as running as root, hard-coding tokens, and running duplicate bot instances.

0
ProgrammingDEV Community ·

SideQuest Streamlines Sideloading on Meta Quest, Opening Path for AmblyoPunch Vision Game

SideQuest has made sideloading apps onto Meta Quest headsets a more direct process, bypassing Meta's official store. The change is aimed at independent developers who previously faced technical hurdles and version-management risks. The article highlights AmblyoPunch, a Quest game by Seven Sportz that applies dichoptic training to help treat amblyopia, a condition affecting about 3% of people. The game runs through four structured phases designed to strengthen the weaker eye and improve binocular fusion and 3D vision.