Prompt Cleaner

The Prompt Cleaner detects and removes copy/paste noise, invisible Unicode characters, encoding artifacts, and optionally selected presentation artifacts from prompt text. It can remove a Unicode BOM, strip zero-width characters, normalize CRLF line endings, remove trailing whitespace, trim outer blank space, collapse excessive blank lines, replace non-breaking spaces, remove suspicious C0/C1 control characters, normalize curly quotes, and normalize em/en dashes. Every change is shown in a cleanup report before you copy.

No. All cleaning runs entirely in your browser using JavaScript. Your prompt text is never uploaded to a server, sent to a language model, or transmitted to any external API. There are no network requests, no cloud processing, and no storage — not even in localStorage. Session history disappears when you close the tab.

A zero-width space (U+200B) is a Unicode character with no visible width that is commonly introduced when copying text from web pages, PDFs, or rich-text editors. When present in a prompt, it can split tokens unexpectedly, inflate the token count, and cause subtle differences in how a language model processes the surrounding text. The Prompt Cleaner detects all zero-width characters — including zero-width space (U+200B), zero-width non-joiner (U+200C), zero-width joiner (U+200D), word joiner (U+2060), and related invisible formatting characters — and reports them before removal.

A Unicode BOM (byte-order mark, U+FEFF) is a special character that appears at the start of a file to indicate the byte order and encoding. It is harmless in a file but can cause problems when pasted into a model context window, because some tokenizers treat it as part of the first token or produce an unexpected output token for it. The Prompt Cleaner detects a BOM at the start of the input and removes it when the "Remove BOM" option is enabled.

C0 control characters are Unicode code points in the range U+0000 to U+001F. Three are expected in normal text: tab (U+0009), line feed (U+000A), and carriage return (U+000D). The remaining 29 are rarely intentional in a prompt and can originate from binary data, null bytes, or corrupted clipboard content. C1 control characters (U+0080 to U+009F) are a Latin-1 supplement range used in legacy encodings. The Prompt Cleaner detects both categories and can remove them with the "Remove C0/C1 controls" option, while always preserving tab, LF, and CR.

The Prompt Formatter handles structural presentation: list marker normalization, ATX heading spacing, and blank-line collapsing. The Prompt Cleaner focuses on encoding hygiene: invisible characters, control characters, BOM detection, replacement-character detection, and line-ending normalization. Both tools preserve code fences and prompt variables, and neither rewrites or restructures your prompt. Use the Formatter when you want consistent presentation; use the Cleaner when you are concerned about invisible or transport artifacts.

The Prompt Optimizer performs heuristic structural analysis — detecting ambiguity, conflicts, missing output format, and section structure, and applying safe transformations like normalization and deduplication. The Prompt Cleaner is purely a noise-removal tool: it only removes characters that are invisible, control-class, or optionally selected cosmetic variants. It never rewrites sentences, reorders sections, or changes the meaning of your prompt.

When the "Preserve protected blocks" option is enabled (which is the default), the cleaner identifies and temporarily removes fenced code blocks (``` and ~~~), inline code spans, bare URLs, Markdown links [text](url), JSON-like blocks, and recognized prompt variables ({{name}}, ${topic}, [UPPERCASE], <variable>). These blocks are restored verbatim after cleaning — no whitespace or character normalization is applied inside them. The inspection report shows how many blocks were protected.

Prompt variables are template placeholders in formats like {{name}}, ${topic}, [OUTPUT_FORMAT], and <variable>. They are intentional parts of a prompt template that should not be altered. If the cleaner modified the spacing or encoding inside a variable, the template might break when it is filled in programmatically. The "Preserve protected blocks" option ensures all recognized variable formats are kept exactly as written.

The replacement character (�) is inserted by decoders when they encounter byte sequences that cannot be interpreted in the expected encoding. It marks the position of lost or corrupted data. The Prompt Cleaner detects replacement characters and reports their count but never removes them — because removing � would delete a marker that indicates something is wrong. The tool explains this clearly in the inspection report and leaves the decision to you.

A non-breaking space (U+00A0) prevents a line break between two words. It is commonly introduced when copying from web pages, Google Docs, or typeset PDFs. In a prompt, NBSP is usually unintentional and may tokenize differently from a regular space (U+0020). The "Replace NBSP with space" option converts all U+00A0 characters to regular spaces. This option is off by default because NBSP is occasionally intentional in multilingual or typographic contexts.

Yes, when the "Normalize line endings" option is enabled (which is the default). The cleaner converts CRLF (carriage return + line feed, \r\n) and bare CR (\r) to LF (\n). This is the safest normalization because LF is the standard line ending for Unix, web, and most modern text processing environments, and because CRLF can occasionally cause tokenization differences when a prompt is pasted from a Windows text editor.

Yes. The cleaner is pure JavaScript with no size limit imposed by a server. Very large inputs (tens of thousands of characters) will process in milliseconds. The file-open button accepts .txt, .md, .markdown, and .prompt files read by the browser FileReader API — no file data is uploaded anywhere.

The inspection report shows a summary chip row with character count, line count, and detected anomaly counts. If invisible characters are found, a table lists each code point, its Unicode name, its category, and how many times it appears. If C0/C1 control characters are detected, a second table lists each one. If replacement characters are present, a warning explains the limitation. If protected blocks were detected, a note shows how many will be preserved.

After you click Clean Prompt, the cleanup report lists every transformation that ran, with the count of instances changed. Each item is color-coded: green for successful cleanups, amber for warnings (such as replacement character detection), and blue for informational items (such as protected blocks preserved). If no changes were needed, the report says so explicitly.

The Swap button copies the cleaned output back into the input field, clearing the output. This is useful when you want to run a second cleaning pass on the result — for example, applying the BOM removal in a first pass and then the NBSP normalization in a second pass with different options selected.

Yes. The cleaner works on any text — system prompts, user turn messages, few-shot examples, or any other prompt component. Encoding artifacts can appear in any pasted text regardless of where it will be used in a conversation.

When the "Normalize curly quotes" option is enabled, the cleaner replaces left and right single quotation marks (‘ U+2018 and ’ U+2019) with the straight apostrophe (') and left and right double quotation marks (“ U+201C and ” U+201D) with the straight double quote ("). This option is off by default because Unicode quotation marks are valid and often intentional. Enable it only if you need pure ASCII punctuation.

When the "Normalize dashes" option is enabled, the em dash (— U+2014) is replaced with two hyphens (--) and the en dash (– U+2013) is replaced with a single hyphen (-). This option is off by default because em and en dashes are semantically distinct from hyphens and are valid Unicode characters. Enable it only if the downstream tool or model context requires pure ASCII punctuation.

No. The Prompt Cleaner removes noise and encoding artifacts — it does not claim that cleaning improves model reasoning, compliance, factuality, or output quality. A clean prompt is easier to audit and has more predictable tokenization, but the tool cannot predict how a specific model will respond to it.

Yes. The Prompt Cleaner is completely free with no registration, no API key, no subscription, and no rate limits. All processing is done in your browser using vanilla JavaScript.

Yes. Once the page has loaded, all cleaning, inspection, copy, and download operations run entirely in your browser with no network calls. You can use it without internet access after the initial page load.

The session history records the last eight cleaning operations in the current browser tab. Each entry shows a preview of the input text, the time, and the number of changes applied. You can click Reuse on any history entry to reload that input into the editor. Session history is memory-only — it disappears when you close or reload the tab and is never written to localStorage or any persistent storage.