Pasted or typed text with invisible/hidden characters is now cleaned and flagged before it's sent, including in fleet composers
What
- Before a prompt is submitted, or before pasted text is sent, Claude Code now runs it through a filter that removes invisible characters, such as bidi marks, zero-width characters, hidden tags, and similar hidden selectors.
- If any such characters are found, sending is blocked. Instead you see a notice like "Removed N invisible character(s) · review and press Enter to send" (or a message that nothing is left to send), and the text is replaced with the cleaned version so you can review it before sending again.
- The same protection now applies to the fleet command bar and the fleet reply composer, not just the main prompt input.
Why
Invisible or hidden characters can be used to hide instructions inside pasted text or make it look different from what it actually contains. Stripping them and asking for confirmation before sending prevents accidentally sending something you didn't intend to.
tengu_tranquil_cloud Not enough to sayNothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.
This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.277: on
These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.
Read once, for one account on one subscription tier, against v2.1.277. It isn't a statement about your account. What a flag value here can and cannot tell you
Whether this behaves differently from the related prompt-submission stripping described elsewhere is not established by the finding.
New in this build: tengu_tranquil_cloud