Sweep 22 Sep 2026 · 15:52Z Build v2.1.280 501 read Stable v2.1.267 Latest v2.1.278 Next v2.1.280 Feeds RSS JSON llms.txt Unofficial
Reading a new release v2.1.280 First look · 1/6 0 findings $0.00 so far
Claude Code v2.1.269 ·

'tee' is now a recognized, sandboxed shell command

The Bash tool now understands the 'tee' command: it's sandboxed, permission-checked, and described in summaries

TierYou'll noticehow much it should matter to you
Useful2my rating, 1 to 5
Signal1worth watching, 1 to 5
AreaBash Toolwhat it touches
KindBug Fixesin v2.1.269,
Group of 4 You'll notice

The Bash tool now understands the 'tee' command: it's sandboxed, permission-checked, and described in summaries

What

  • tee is now implemented as a supported shell builtin in the sandbox, piped through the existing sha*sum-style handler, and classified as a write permission; special device targets (/dev/null, /dev/stdout, /dev/stderr, /dev/tty) are filtered out of its arguments before permission checks run.
  • The command path-validator now special-cases tee: a tee call with no target paths passes immediately ("Path validation passed for tee command") instead of going through the usual working-directory restriction logic.
  • The basename-normalizing helper used to detect file-affecting shell commands (previously only rm/rmdir) now also matches tee, case-insensitively and with a .exe suffix stripped.
  • tee was added to the bash-command-summary descriptions, described as "write to files in".

Why Previously tee wasn't recognized as a file-writing command, so it could bypass the usual write-permission and working-directory checks or produce confusing summaries; now it's treated consistently with other file-writing commands.

Read from
Names in the bundletee
Since it was published

The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.

Added since A small documentation edit on browser-use-tool touched a line naming tee after this was published. Anthropic has trained the model to resist these prompt injections and has added an extra layer of defense. If you use the browser use tool, classifiers will automatically scan what the browser returns, such as page text or screenshots, to … agents-and-tools/tool-use/browser-use-tool see the edit
Confirmed since Anthropic's documentation has since written up tee, on Use Claude Code on the web. Open the session on claude.ai or the Claude mobile app to check progress or interact directly. From there you can steer Claude, provide feedback, or answer questions as in any other conversation. claude-code-on-the-web see the edit
How sure we are
One source agreesOne thing we can check says the same as this entry.
Anthropic's documentation agreesAnthropic's documentation has since written up tee, on Use Claude Code on the web.

See this entry in the whole of v2.1.269 →