The inspector flags known invisible and directional code points without changing indentation, line endings, identifiers, string contents, or formatting on its own.
Source code inspection
Does Claude Code Have Watermarks?
Check pasted source code for invisible Unicode and formatting controls that can confuse review, parsing, or display. Cleanup stays explicit because some characters are meaningful in code and strings.
Analyze + Clean stay local Safe fixes selected in text mode No fake detection score
Free Claude text cleaner
Paste your text to check it instantly
Your text stays in this browser
We check for invisible characters and unusual spacing. Your words are not rewritten.
0 charactersInstant check
Formatting issues found
Safe to clean
Low-risk formatting changes only
- 0 hidden characters
- 0 unusual spaces
Needs your review
Left unchanged by safe clean
No formatting issues found
Your text is ready to copy. This check covers known Unicode and spacing artifacts.
Nothing safe to clean
Advanced details Code points and exact locations
Before / After
Review what changed
Only the items you selected were changed. Your sentences were not rewritten.
BeforeVisible markers added
AfterComplete cleaned text
FreeNo accountRuns locally in your browser
Developer boundary
Code hygiene findings are not code-authorship evidence
Exact Unicode locations can support review and debugging. Anthropic has not published code-specific detection details, so this page does not infer authorship from formatting artifacts.
Finding a zero-width or direction character does not prove Claude wrote the code. Finding none does not prove the code lacks a model-level statistical mark.
What Anthropic says about Claude Code output
Anthropic’s current guidance says markings cover output from supported models wherever Claude is used, including Claude Code, and that embedded watermarks apply to all generated text. The company has not published code-specific technical detection details.
Code hygiene is a separate, verifiable layer
Zero-width characters, bidirectional controls, byte order marks, and unusual spaces can affect review, rendering, parsing, or security analysis. This inspector reports those concrete code points without automatically changing indentation, line endings, identifiers, or string contents.
Review context before removing a character
A flagged character may be intentional inside a test fixture, string literal, comment, regular expression, identifier, or multilingual file. Inspect the line and column, review the diff, and run the repository’s formatter, type checker, and tests after cleanup.
What the result cannot establish
A Unicode finding cannot prove Claude authorship. A clean Unicode report cannot prove that code lacks a model-level mark. Short snippets and constrained syntax may also leave too little material for reliable detection, according to Anthropic’s general limitations.
Clear answers, narrow claims
Frequently asked questions
Does Claude Code have watermarks?
Anthropic says markings cover output from supported models across Claude Code and that embedded watermarks apply to all generated text. It has not published a code-specific detection mechanism.
Does Claude watermark source code specifically?
Anthropic’s current guidance includes Claude Code output but does not explain how detection behaves for particular programming languages, short snippets, or highly constrained syntax. This tool does not fill that gap with guesses.
Can hidden Unicode prove that Claude wrote code?
No. Invisible characters can come from editors, copy-and-paste, dependencies, encoding, or deliberate source text. Their presence is a code-review signal, not proof of authorship.
Should I remove every flagged character from code?
No. A character may be intentional inside a string, comment, identifier, fixture, or multilingual source file. Review its location, inspect the diff, and run the project’s normal tests after any cleanup.