Clean user-facing text
Final hygiene pass on prose the user owns or is authorized to process. Unicode cleanup is deterministic. Statistical-watermark reduction is best-effort. Never claim a rewrite proves human authorship or is undetectable. Keep required academic, legal, platform, and regulatory disclosures.
Workflow
- Identify the prose that readers will see.
- Protect non-prose spans:
- fenced and inline code
- commands, paths, URLs, identifiers, API names, and exact values
- formulas, citations, and text the user asks to quote verbatim
- Preserve every claim, fact, number, name, citation, and requirement.
- Establish the writing brief before changing prose:
- use a voice sample only when the user owns it or is authorised to use it; don't imitate another named person
- when there is no sample, make the prose clear and natural without pretending to imitate a particular person
- keep required disclosures, uncertainty, and the writer's actual point of view
- Rewrite the remaining prose once:
- vary clause order, sentence boundaries, rhythm, connectors, and function words
- replace formulaic transitions and filler with direct, natural wording
- keep the concrete details and judgement that make the text recognisable as the writer's
- treat unusual grammar, repetition, directness, and phrasing as possible voice or accessibility choices; change them only when the user asks or when they create a clear reading problem
- preserve the requested language, tone, structure, and formatting; never translate unless asked
- for non-English text, use fluent constructions native to that language rather than English sentence patterns
- do not add or remove claims merely to increase variation
- For text artifacts or supplied text files, run the deterministic Unicode pass after rewriting.
- Return only the polished result unless the user asks for an audit or explanation.
For practical guidance on preserving a writer's voice and removing formulaic prose, read references/writing-in-your-voice.md whenever the user asks to retain or adjust voice.
Deterministic Unicode pass
Resolve SCRIPTS to this skill's scripts/ directory.
Use the available Python 3 launcher for the platform. Replace PYTHON below
with python3 on most macOS/Linux systems, py on Windows, or another verified
Python 3 command.
Inspect first when editing an existing file:
PYTHON "$SCRIPTS/inspect_text.py" --json INPUT
PYTHON "$SCRIPTS/clean_text.py" INPUT -o OUTPUT --stats --no-normalize-spaces
PYTHON "$SCRIPTS/inspect_text.py" --json OUTPUT
Use - for stdin. Prefer a new *.cleaned.* output unless the user explicitly requests in-place editing.
Use --no-normalize-spaces by default so NBSP, narrow no-break spaces, figure spaces, and CJK ideographic spaces retain their layout semantics. Normalize spaces only when the user requests it.
Do not use --aggressive-homoglyphs, --nfkc, or --strip-emoji-glue unless the user requests aggressive normalization and accepts possible changes to multilingual text, emoji, directionality, or typography.
The scripts support plain text, source text, Markdown, and HTML source as text. For mixed Markdown or HTML, inspect hit positions first. If a hit falls inside protected code, attributes, or another non-prose span, do not run whole-file cleanup; clean only the prose segments or leave that hit unchanged. Do not pass binary containers such as PDF, DOCX, images, or archives.
For a chat-only response that is not written to a file, perform the rewrite workflow directly. Do not claim that the chat response received a deterministic post-send Unicode filter.
Code boundary
When prose and code are mixed, rewrite prose only. Never rename variables, alter string literals, reformat code, or change executable output as part of this skill. If a Markdown or HTML file contains executable snippets, preserve those spans byte-for-byte whenever practical.
Reporting
When the user asks for an audit, distinguish:
- Verifiable: Unicode characters removed or replaced, with script counts.
- Best-effort: prose rewritten to alter token and syntax patterns.
- Not established: official detector evasion, human authorship, or removal of a vendor's secret-key watermark.
For technical background, read references/watermark-notes.md. For misuse or disclosure questions, read references/responsible-use.md.
1---2name: clean-user-facing-text3description: Audit and finalize authorized natural-language text meant for readers: strip suspicious invisible Unicode, then rewrite prose while keeping facts, meaning, and the writer's voice. Use when the user asks to clean, humanize, polish, or finalize articles, manuscripts, reports, documentation, emails, product copy, UI text, Markdown, or HTML prose, or when a project rule explicitly requires this workflow. Don't use for code-only tasks or undisclosed authorship evasion; leave code, commands, identifiers, paths, APIs, formulas, citations, required disclosures, and verbatim quotations unchanged.4---5
6# Clean user-facing text
7
8Final hygiene pass on prose the user owns or is authorized to process. Unicode cleanup is deterministic. Statistical-watermark reduction is best-effort. Never claim a rewrite proves human authorship or is undetectable. Keep required academic, legal, platform, and regulatory disclosures.
9
10## Workflow
11
121. Identify the prose that readers will see.
132. Protect non-prose spans:
14 - fenced and inline code
15 - commands, paths, URLs, identifiers, API names, and exact values
16 - formulas, citations, and text the user asks to quote verbatim
173. Preserve every claim, fact, number, name, citation, and requirement.
184. Establish the writing brief before changing prose:
19 - use a voice sample only when the user owns it or is authorised to use it; don't imitate another named person
20 - when there is no sample, make the prose clear and natural without pretending to imitate a particular person
21 - keep required disclosures, uncertainty, and the writer's actual point of view
225. Rewrite the remaining prose once:
23 - vary clause order, sentence boundaries, rhythm, connectors, and function words
24 - replace formulaic transitions and filler with direct, natural wording
25 - keep the concrete details and judgement that make the text recognisable as the writer's
26 - treat unusual grammar, repetition, directness, and phrasing as possible voice or accessibility choices; change them only when the user asks or when they create a clear reading problem
27 - preserve the requested language, tone, structure, and formatting; never translate unless asked
28 - for non-English text, use fluent constructions native to that language rather than English sentence patterns
29 - do not add or remove claims merely to increase variation
306. For text artifacts or supplied text files, run the deterministic Unicode pass after rewriting.
317. Return only the polished result unless the user asks for an audit or explanation.
32
33For practical guidance on preserving a writer's voice and removing formulaic prose, read `references/writing-in-your-voice.md` whenever the user asks to retain or adjust voice.
34
35## Deterministic Unicode pass
36
37Resolve `SCRIPTS` to this skill's `scripts/` directory.
38Use the available Python 3 launcher for the platform. Replace `PYTHON` below
39with `python3` on most macOS/Linux systems, `py` on Windows, or another verified
40Python 3 command.
41
42Inspect first when editing an existing file:
43
44```bash
45PYTHON "$SCRIPTS/inspect_text.py" --json INPUT
46PYTHON "$SCRIPTS/clean_text.py" INPUT -o OUTPUT --stats --no-normalize-spaces
47PYTHON "$SCRIPTS/inspect_text.py" --json OUTPUT
48```
49
50Use `-` for stdin. Prefer a new `*.cleaned.*` output unless the user explicitly requests in-place editing.
51
52Use `--no-normalize-spaces` by default so NBSP, narrow no-break spaces, figure spaces, and CJK ideographic spaces retain their layout semantics. Normalize spaces only when the user requests it.
53
54Do not use `--aggressive-homoglyphs`, `--nfkc`, or `--strip-emoji-glue` unless the user requests aggressive normalization and accepts possible changes to multilingual text, emoji, directionality, or typography.
55
56The scripts support plain text, source text, Markdown, and HTML source as text. For mixed Markdown or HTML, inspect hit positions first. If a hit falls inside protected code, attributes, or another non-prose span, do not run whole-file cleanup; clean only the prose segments or leave that hit unchanged. Do not pass binary containers such as PDF, DOCX, images, or archives.
57
58For a chat-only response that is not written to a file, perform the rewrite workflow directly. Do not claim that the chat response received a deterministic post-send Unicode filter.
59
60## Code boundary
61
62When prose and code are mixed, rewrite prose only. Never rename variables, alter string literals, reformat code, or change executable output as part of this skill. If a Markdown or HTML file contains executable snippets, preserve those spans byte-for-byte whenever practical.
63
64## Reporting
65
66When the user asks for an audit, distinguish:
67
68- **Verifiable:** Unicode characters removed or replaced, with script counts.
69- **Best-effort:** prose rewritten to alter token and syntax patterns.
70- **Not established:** official detector evasion, human authorship, or removal of a vendor's secret-key watermark.
71
72For technical background, read `references/watermark-notes.md`. For misuse or disclosure questions, read `references/responsible-use.md`.
majiayu000