Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-09-18 · high
Claude Code's sandboxed Bash on Linux returned exit code 0 for failed commands under zsh, so failed builds and tests looked clean.
Why now: The fix is in v2.1.275, covered here as of September 18, 2026.
Sandboxed Bash on Linux was returning exit code 0 for commands that failed, when the shell was zsh, until Claude Code v2.1.275 fixed it.
The model reads the exit code to decide whether a step worked. A failed build, test, or migration in that setup looked clean, and the agent kept building on it.
The same release fixes a second bug in the undo path. Run in a forked or background session, /rewind could restore a zero-filled or truncated file when the file-history backup hadn't finished copying. That puts file corruption right where someone has already discarded their working state and is counting on the restore to bring it back.
Neither bug shows up when someone is watching the terminal. Sandboxed execution on Linux and background or forked sessions are both setups where nobody watches every command scroll by. Not watching is what makes those workflows useful in the first place. If you run Claude Code in a sandbox on Linux, or use background sessions with /rewind, updating to v2.1.275 closes both gaps.
The release notes don't say how long the exit-code bug was present or how many Linux sandbox runs it touched before the fix.
Each link below shares sources, entities, or timing with this story.
One Claude Code release fixed two independent permission-check bypasses on the same day. That's the story. Version 2.1.221, shipped August 4, patches a Bash tool bypass where zsh could execute hidden commands embedded inside [[ ]] regex conditionals. The approval prompt never...
Each project gets a throwaway VM where the agent has full tool access to Docker, git, compilers and package managers with no path back to the host (GitHub). coop setup installs Firecracker and a guest kernel on Linux; macOS goes through Lima via Homebrew. Usage is coop up then...
+7,546 stars this week for an "Application Development Environment" running each concurrent agent in its own worktree. Works with any terminal CLI agent — Claude Code, Codex, OpenCode, Pi, Cursor, Copilot, Grok, 30+ others — across macOS, Windows, Linux, iOS via App Store/Test...
The most quotable line in a Linux kernel commit this year: "a debug session from hell, enormously helped by an AI doing much of the grunt-work." Torvalds landed drm/xe: Don't hand out the flat CCS storage as usable VRAM on August 22. The commit message is the artifact. He note...
The agent-security topic holds 42 repos above 100 stars, four from large companies rather than startups: NVIDIA/SkillSpector (14,498 stars, scanning Claude Code/Codex/MCP skills for prompt injection), Tencent/AI-Infra-Guard (4,467, red-teaming with Many-Shot/PAIR/GOAT/ActorAtt...
msitarzewski/agency-agents added 446 stars today, packaging personas across "divisions" (frontend specialists, community experts, fact-checkers, reality checkers), each defined with a voice, a process, and concrete deliverables rather than a generic prompt template (GitHub). I...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.