A single daily edition of announcements from the AI coding tools you use, read from a curated source list, summarised, and linked back to the original. No infinite feed, no account, no advertising.
The hidden directories in your repo are where a credential stealer now hides
Google's Threat Intelligence Group reports that AI-assisted coding tools are now a primary target for threat actors, tying them to several large-scale software supply chain compromises in 2025 and early 2026. The financially motivated group UNC6780 hit PyPI, npm and Docker Hub, placing malicious files in hidden project directories to blend into developer noise, and used the Dustmaker credential stealer to pull tokens from GitHub Actions runners. One command-and-control server held over 23,800 harvested secrets, and a separate mass harvesting campaign ran in under six hours.
A plugin path with a backslash slipped past the symlink containment check
Claude Code 2.1.265 fixes a plugin path containing a backslash bypassing the symlink containment check on macOS and Linux, ten days after 2.1.251 shipped the containment work that check belongs to. The same release stops plugin directories whose names begin with two dots being wrongly refused as outside the plugin root, adds a 1 GB cap on tool results saved to disk, and lets --plugin-dir point at a folder where every child carrying a manifest loads, including ones added while it runs. A follow-up the same evening repairs a regression affecting LLM-gateway and proxy setups.
A 260K-context model on OpenRouter returns over a thousand tokens a second
Inception's Mercury 2.5 is a diffusion model running at 1,107 tokens per second on widely available NVIDIA GPUs, and the company puts it 40% above Mercury 2 on intelligence, comparable to cost-optimised frontier models. List price is $0.20 per million input tokens and $0.75 per million output, discounted 80% at launch to $0.04 and $0.15. It is reachable through the Inception API with 100 million free tokens, Baseten, OpenRouter or the web chat, and the company names coding subagents and assistants as a target workload.
An enterprise policy now decides what the JetBrains agent may read and reach
Administrators can centrally configure sandbox behaviour for GitHub Copilot in JetBrains IDEs, shipping as a public preview. Managed policies cover sandbox enablement, filesystem and network access, proxy settings, developer-tool access and macOS Keychain access, and managed restrictions take precedence over user settings, with the IDE locking the affected controls. Turning it on requires the organisation to enable the Editor Preview feature flag or configure managed settings.
One skill file stops your agent from burying the answer under preamble
The i-have-adhd skill enforces ten rules on a coding agent's replies: lead with the next action, number multi-step tasks, cap lists at five items, and end with one concrete next step. It installs as a plugin from the repository or through its INSTALL.md, and ships configuration files for Claude, Gemini, Kimi and Qwen. The repository carries 31.6k stars and reached the Hacker News front page today.
