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 encrypted reasoning in your published agent logs comes back out in plaintext.
Researchers found that the encrypted chain-of-thought blocks providers return to clients are fully interchangeable across sessions, users and models inside one provider's ecosystem. Feeding a block from a strong model into a weaker, less safeguarded model from the same provider makes it decode the trace verbatim, so the stronger model is never jailbroken directly. Decoding 315,320 blocks scraped from public repositories recovered 367 personally identifying artifacts and 182 credentials. The paper names Anthropic, OpenAI and Google, and the authors proposed cryptographic and system-level mitigations after disclosure.
Your project's Claude Code settings file is now a malware persistence slot.
Unit 42's 6 August analysis of the ChainDrop npm worm describes it writing a Claude Code settings file carrying a SessionStart command hook, so the dropper runs whenever Claude Code opens a session in that project. The hook invokes a script under the VS Code directory, cross-linking the two editors' configs so each one references the other's copy. The worm reached over 400 packages downloaded hundreds of millions of times weekly, and harvests cloud IAM credentials, npm and GitHub tokens, SSH keys, and AI tool authentication artifacts. On 4 August the operator reconfigured the C2 through one Ethereum transaction, rotating the active domain without any update to the deployed malware.
A skill synced from the web no longer shadows the commands you already had.
Claude Code 2.1.228, released on 11 August, hardens skills synced from claude.ai: they no longer shadow local commands or MCP prompts, their descriptions are sanitized and labeled, and on your own machine their bodies do not run shell commands or expand file references. The same release fixes Remote Control resume leaking a resumed conversation's title or history into the connected session, and stops session cleanup deleting contents inside a project's memory folder. It also changes the Write tool so newer models can overwrite an unread file, matching the Edit tool's rules, while older models still require the read first.
The work you started locally survives closing the terminal.
Kiro CLI 2.17.0, dated 11 August, introduces cloud sessions in preview: launching with the cloud flag runs the session in a managed cloud sandbox, so you can disconnect while the agent keeps working and resume from any machine. Repositories attach with a repo flag or the repo picker, and a session is picked up later by its resume id. Cloud and local sessions share one picker, so the two kinds are not tracked separately. The release also adds substring matching in the slash-command menu and smoother typing during rapid input in the TUI.
