Maya: Hey builders! Here's what's on the radar today on Claude Code Review. Anthropic's handing out cloud credits, a 2.1.291 patch, a sub-agent that wiped a home folder, the Senzu safety hooks, and Claude Code mods. I'm Maya.

James: And I'm James. Let's get into it.

Maya: So Anthropic is giving existing Claude Code subscribers a one-time cloud credit: one hundred dollars on Pro, two hundred fifty on Max. You claim it by October seventh, and it only pays for cloud sessions.

James: Right, and what a cloud session actually is matters here, Maya. Claude Code runs on a virtual machine Anthropic manages, clones your GitHub repo, works on its own branch, and hands back a pull request. That's why work survives a closed laptop.

Maya: Exactly right. To claim it, type slash claim dash credit in Claude Code, or use the banner on the Usage page. Eligibility is narrow: you needed an active Pro or Max subscription when it began on September twenty third.

James: And one catch worth flagging. The credit doesn't cover local Claude Code, chat, or Cowork, and unused credit disappears November fourth. The walkthrough author suggests trying one task early, something that normally takes half a day locally.

Maya: Onto the patch. Claude Code two point one point two nine one fixes two regressions. The first, from two point one point two nine zero, meant cloud sessions could drop answers to permission prompts, so what you decided wasn't reliably carried through.

James: That's the one that would bite you, Maya. The second bug, from two point one point two eight eight, could lose a session's last messages on quit. So anyone on two eight eight through two nine zero should update now.

Maya: Exactly. The release notes are just two lines, no root cause, no frequency. But a separate commentary post used it to argue that a permission answer being given doesn't prove it was preserved until the action executed.

James: This next one is hard to read. A user on Windows eleven says a cleanup sub-agent wiped about one hundred sixteen gigabytes from their home directory. It issued rm dash rf against a short-name alias that points at the full home folder.

Maya: And nothing resolved that alias before the delete proceeded, James. Then two failures stacked: the delete exceeded the foreground timeout, moved to background, and the agent reported it stopped successfully while the process kept deleting files alphabetically for roughly fifty more minutes.

James: Right, and its confirmation came from a shallow top-level listing, which still shows folder names while contents vanish underneath. Losses were broad: a multi-month Kaggle portfolio, forty projects, toolchains, plus credentials like dot ssh and the GitHub CLI login.

Maya: It's one report from one machine, filed as GitHub issue nine nine one nine three, open with no maintainer response at the time of writing. But it points at two places: path handling before a destructive command, and confirming a stopped process is actually dead.

James: Which is a clean segue to Senzu. It's a free, open-source package for Claude Code and Codex that bundles sixteen hooks. A hook is a script that runs when the agent tries something, and in Senzu's case it can refuse the action and explain why.

Maya: And the argument is reliability, James. The developer says an agent reads a CLAUDE dot md file and almost always follows it, which isn't enough. So the guard hook blocks git push, rm dash rf, and deploying to production without an explicit variable plus approval.

James: Exactly, and secrets guard recognizes AWS, Stripe and OpenAI keys by shape so they're never written into code. Setup is two commands: plugin marketplace add, then plugin install senzu dash all. It's MIT-licensed and written in Spanish on purpose.

Maya: Last one, and it's lighter. Claude Code mods are small add-ons that change how the interface looks or behaves, and a handful of guides now show where to start, from a safety check to a pet above the prompt.

James: And the practical starting points are good, Maya. One builder suggests linking two official Anthropic mods in your prompt as patterns. Another article walks through a twenty-line mod that stops rm dash rf, pitched as a better first mod than cosmetic spinners.

Maya: And the fun one: a free open-source mod puts a Pokemon above the prompt. It walks around while Claude works, drops a pokeball per subagent, and gains XP from your turns. The developer says Claude wrote the whole thing, in the claude dash pokemon dash mod repository.

Maya: That's your Claude Code Review daily brief. Check claudecodereview.ai for the full stories. Keep building.