Claude's Banning Users and Everyone's Confused

"Getting banned from Claude for creating a Claude.md file is the 2025 equivalent of getting kicked out of a restaurant for reading the menu too carefully."

Hello Builders! 👋

A developer got banned from Claude for creating a Claude.md file – you know, those scaffolding files that tell AI assistants how to work with your codebase. The irony is thick enough to cut with a knife: using Claude's own recommended practice triggered their overzealous content policy system.

This sparked a 559-comment firestorm on HN because it hits a nerve every builder feels – we're building on platforms that can pull the rug out with zero warning. The automated moderation flagged legitimate dev work as policy violation, and getting unbanned requires navigating Kafka-esque support channels.

Petri 2.0 Ships for Alignment Audits - Anthropic's open-source alignment testing tool gets upgraded after research groups actually started using it in production.

Injective Integrates Claude Code - Web3 platform adds Claude directly into their dAppBuilder, because why hire developers when you can prompt them.

Pochi Ships Plan Mode - New planning feature hints at modular architecture letting you swap AI capabilities like Lego blocks.

Skip Goes Open Source - The Swift-to-Kotlin transpiler drops paid tiers and opens everything up for free.

5x Better Agentic Coding - Few-shot prompting techniques that actually move the needle on LLM coding performance with real benchmarks.

Prompt Sophistication = Output Quality - Anthropic research proves what we suspected: better prompts get dramatically better responses, backed by data.

Free RAG & MCP Course - Hands-on crash course covering RAG fundamentals and Model Context Protocol from scratch.

Fighting Spam with Gemini - DEV Community's practical implementation of LLMs for content moderation at scale.

AI Coding's Groundhog Day Bug - You fix the AI's bug, then it cheerfully reintroduces the exact same bug next iteration.

Traditional Engineers Watch Vibe Coders - The schadenfreude is real as classically-trained devs watch prompt engineers discover technical debt the hard way.

eBay Bans AI Shopping Agents - Platform explicitly blocks "buy for me" AI agents in updated TOS, 341 HN comments discuss implications.

AI Agents Fail Workplace Tests - New benchmark testing models on real consulting, banking, and legal tasks shows most agents aren't ready for primetime.

a16z Drops $150M on Inferact - The vLLM maintainers just raised a massive seed round to keep the open source project alive and kicking.

iOS App Store Surges 60% - After three years of flatline growth, new app releases exploded in December thanks to AI-assisted coding making everyone a builder.

AI Writing Nearly 1/3 of US Code - Up from just 5% in 2022, which explains why your PRs suddenly look like everyone learned the same coding style.

Mobile Apps Outearned Games in 2025 - AI apps drove consumer spending past gaming for the first time, proving people will pay for ChatGPT wrappers apparently.

Blockit Raises $5M for Calendar AI - Former Sequoia partner builds AI agent that negotiates meeting times directly with other calendars, Sequoia leads seed round.

Humans& Builds Coordination Models - Ex-Anthropic/OpenAI/DeepMind crew raises funding to build foundation models for collaboration instead of just chat.

Postgres Rules for Agents - Supabase dropped best practices for AI agents hitting Postgres, covering query performance, connections, and RLS security.

ATS-Beating Resume Builders - Five open-source tools designed to survive the 75% of resumes that get auto-deleted before humans see them.

AI Beat Our Hardest Interview - Anthropic's performance engineering exam was legendary until Opus 4.5 solved it—here's their new approach.

Wild that Anthropic's dropping a new constitution the same week someone gets banned for a Claude.md file. The vibes are definitely shifting from "move fast" to "move carefully with 50 pages of guidelines."

What are you shipping this week? Any Claude.md casualties in your codebase?

Keep shipping,

P.S. If you're one of those engineers watching everyone else vibe code apps, maybe it's time to admit the vibes are working. Or at least grab some popcorn.