prose-check
Skill4 starsChanged 19 days agoMITView on GitHub
Write and review texts, in Spanish or English, that do not read as model-generated. Gives the concrete markers that give away an AI text —negative parallelism, rule of three, epigrammatic headlines, em dashes, recurring vocabulary, grandiloquence—, a command that counts them on the real text, and the method for rewriting without falling into them. Use when writing any text another person will read: documentation, presentations, commit messages, emails, proposals, posts and interface copy, and always before delivering a document to a client. Also when someone says a text "suena a IA" or "sounds like AI". Formerly called redaccion.
In this repository46 other files it gives its agents.
- AGENTS.mdAGENTS.md
- add-repo-to-docsSkill
- adrSkill
- auth-bypass-localSkill
- before-afterSkill
- bug-huntSkill
- carry-envSkill
- code-healthSkill
- crux-doctorSkill
- crux-initSkill
- css-to-componentsSkill
- deploy-fallbackSkill
- deploy-watchSkill
- device-loopSkill
- drive-browserSkill
- e2eSkill
- end-sessionSkill
- eval-fixturesSkill
- explain-planSkill
- flowSkill
- grillSkill
- merge-conflictsSkill
- motion-videoSkill
- oauth-reauthorizeSkill
- prod-signalsSkill
- promote-to-docsSkill
- read-documentSkill
- render-loopSkill
- repo-mapSkill
- review-prSkill
- root-causeSkill
- runbookSkill
- security-auditSkill
- share-localSkill
- shipSkill
- skill-auditSkill
- skill-creatorSkill
- teachSkill
- ui-behaviourSkill
- ui-tellsSkill
- visual-diffSkill
- visual-testSkill
- wizardSkill
- workspaceSkill
- write-testsSkill
- x-postsSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
When: before delivering a text to an outsider.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
prose-check
A generated text is recognised by its shape before its content. The markers are
few, concrete and counted with a command. More examples and the reason behind
each rule: <this-skill-directory>/references/why.md.
When: before delivering a text to an outsider; when writing documentation, a presentation, a proposal or a long email; when someone says a text "suena a IA".
It works on Spanish and English text. The examples below come in both languages because they are the data it looks for.
The seven markers
The first three are enough to recognise a generated text.
- Negative parallelism. «No es X, es Y», «no se trata de X, sino de Y»; "it's not X, it's Y", "not just X, but Y". → Drop the negation and keep what you assert: «Un plan sirve para que alguien lo pare a tiempo».
- Rule of three. Three adjectives or phrases where the content called for one or two («claro, sencillo y directo»; "fast, simple and reliable"). → If the third adds nothing, it is surplus.
- Epigram as headline. «Lo que toda pieza tiene que cumplir» instead of «Requisitos»; "What makes a good test" instead of "Tests". → The standard term of the trade.
- Excess em dashes, where a comma, a parenthesis or a full stop belonged. → One or two per document.
- Aphoristic colons to finish with a maxim («El resultado es claro: no funciona»; "The answer is simple: measure it"). → Full stop, or a single assertion.
- Recurring vocabulary. Spanish: «clave», «fundamental», «crucial», «robusto», «sólido», «potente», «profundizar», «panorama», «cabe destacar», «garantizar», «optimizar», «impulsar». English: "delve", "crucial", "pivotal", "robust", "seamless", "leverage", "landscape", "foster", "holistic", "unlock", "harness", "streamline", "it's worth noting". → The concrete verb or noun: «permite mejorar la legibilidad» → «se lee mejor»; "leverages caching to optimize" → "caches".
- Grandiloquence: linking something to a bigger theme or closing with a reflection that has no data («Esto es lo que separa un dato de una afirmación»; "and that makes all the difference"). → Delete it.
How to review
crux prose check <file…> # counts
crux prose check --text "…" # one string
crux prose detail <file…> # each case with its line
crux prose check --lang es <file> # force the language; detected otherwiseThe command counts and locates; deciding whether each case is surplus is up to you. Markers 1, 2, 3, 6 and 7 score; the em dash and the colons are reported apart, because a pattern cannot tell a punchline from an explanation. It exits 1 when the text reads as generated.
Indicative threshold for about 2,000 words: zero negative parallelisms, fewer than three rules of three, no epigrammatic headline.
What it leaves out on purpose
- Quotes, «quoted text», “quoted text” and code: a text that teaches what to avoid has it written out.
- «clave» as a noun (una clave SSH); it only counts as an adjective.
- «lo que hace falta», which contains «lo que hace».
- Colons that open an enumeration.
What it does not fix
Register: a text can pass all seven and be useless because it uses jargon its reader does not know. Before reviewing the form, ask yourself who reads it. And do not take it to the extreme: a well-placed dash or an enumeration of three real things is good writing.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

