agentleFS
izkrenyizkreny/agentifico/skills/review-text/SKILL.md

review-text

Skill4 starsChanged 13 days agoMITView on GitHub

Review prose for spelling, grammar, punctuation, style, clarity, tone and readability, reporting the issues or applying minimal edits when asked. Use for proofreading, or tightening the phrasing of, selected text, markdown, documentation, prompts, AGENTS.md, README files, code comments, UI copy, emails and commit messages. Invoked explicitly only, by the user typing `/review-text`: an agent never loads it on its own. Not for reviewing code logic or a pull request, translating, or rewriting text in a new voice.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 8 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Only fix edits the text, because an edit the user did not ask for is one they have to find and undo.

    line 33
  2. Distinctive:

    Read the exact current text before making any claim about it, because a claim about text you have not read is a guess.

    line 37
  3. Distinctive:

    Treat a deliberate rule-break that carries the author's voice, such as a fragment for emphasis, as their choice rather than an error, …

    line 43
  4. Distinctive:

    Leave code, identifiers and quoted literals alone unless the user asks, because a changed identifier breaks whatever refers to it.

    line 45
  5. Distinctive:

    Keep the formatting and markdown structure unless the user asks for another or the structure itself is the problem, because readers and …

    line 46
  6. Distinctive:

    Remove filler phrases, duplicated words and unnecessary commas, because each costs the reader attention and carries nothing.

    line 57
  7. Distinctive:

    Keep parallel list items grammatically parallel, because a list that changes shape midway reads as two lists.

    line 58
  8. Very specific:

    Keep plain words plain: use stays use rather than utilize, and people stays people rather than individuals, because a formal swap changes …

    line 59
  9. Distinctive:

    Ask before any change that would alter the meaning, because only the author knows which meaning was intended.

    line 65
  10. Distinctive:

    Keep every fact, name, number, date, quote and citation as the source gives it, and each claim at the scope and certainty it was written …

    line 66

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Review text

When to Use

  • Proofread selected text or a whole file for spelling, grammar, punctuation, style or clarity.
  • Review markdown, documentation, prompts, AGENTS.md files, code comments, UI copy, emails or commit messages.
  • Tighten phrasing without changing meaning.

Arguments

The arguments are a target, then an optional scope, then an optional action. A last word of fix is the action, a last remaining word of issues or style is the scope, and everything before them is the target; to review text that ends in one of those words, quote the text. The target is the text to review, or the path of a file holding it. With no target, review the active selection or the active file where the harness exposes them, and otherwise ask, because a review of a guessed target wastes the whole pass.

issues and style set the scope, and fix is the one action; without it the skill reports, which is the default:

ArgumentWhat it does
issuesReports spelling, grammar and punctuation errors.
styleReports wording, structure, ambiguity, tone and readability suggestions.
fixApplies the corrections for the stated scope: issues, style, or both when no scope is stated.
noneReports what issues and style find, and fixes nothing.

Only fix edits the text, because an edit the user did not ask for is one they have to find and undo.

Instructions

  • Read the exact current text before making any claim about it, because a claim about text you have not read is a guess.
  • Before judging style, look for the project's own style guide, such as a STYLE.md file, a Vale configuration or the style rules in its AGENTS.md. The user's instructions for this request win over it, and it wins over this skill's ## Style Heuristics, because the project's conventions are the author's and the heuristics are only defaults.
  • Give every finding one level, because the level tells the user how much weight it carries: error for text that is wrong in any register, such as a misspelling or a broken agreement; warning for text that is very likely wrong but has a deliberate reading; suggestion for a preference, which is always optional. A choice you would not make is a suggestion at most, because different is not wrong.
  • Report a weak pattern, such as one passive sentence, only where it recurs or clearly costs the reader, because a finding on every instance buries the ones that matter.
  • Say no issues found when there are none, because an invented finding costs the user a check and teaches them to distrust the rest.
  • Preserve the author's tone, an informal tone included, unless the user asks for another, because the text is theirs and the review serves it.
  • Treat a deliberate rule-break that carries the author's voice, such as a fragment for emphasis, as their choice rather than an error, because smoothing it out erases the voice the rule-break was there for.
  • Make the smallest edit that fixes each issue and change nothing else in the text, because every extra change is one more line the user has to check.
  • Leave code, identifiers and quoted literals alone unless the user asks, because a changed identifier breaks whatever refers to it.
  • Keep the formatting and markdown structure unless the user asks for another or the structure itself is the problem, because readers and tools may depend on it.

Response Pattern

  • For a report, say whether anything was found, then list the findings error first, warning next and suggestion last, and by consequence within each level, so the one that matters most is read first. Give each the quoted original, a short explanation and the suggested wording. Add a cleaned-up snippet only where the list alone is hard to apply, because otherwise it repeats the list.
  • For an edit, make the edits, then show a numbered table of them with the columns #, Original, Revised and Why, so the user can check each change without diffing the file and undo one by its number ("revert 3"). For inline text, return the corrected text in full before the table, because there is no file for the user to open.

Style Heuristics

  • In agent and configuration files, prefer direct instructions over biography, because an agent acts on an instruction and can do nothing with a backstory.
  • In instruction files, prefer operational verbs such as apply, use, follow, load and check, because each names an action the reader can take.
  • Remove filler phrases, duplicated words and unnecessary commas, because each costs the reader attention and carries nothing.
  • Keep parallel list items grammatically parallel, because a list that changes shape midway reads as two lists.
  • Keep plain words plain: use stays use rather than utilize, and people stays people rather than individuals, because a formal swap changes the voice and adds nothing.
  • Judge a style rule by the text's audience, because the right choice can flip with it: extensible beats extendable in software writing and loses to it elsewhere.
  • In instruction files, suggest stating the behaviour wanted rather than only the behaviour banned, because a prohibition puts the banned behaviour in front of the agent that reads it.

Boundaries

  • Ask before any change that would alter the meaning, because only the author knows which meaning was intended.
  • Keep every fact, name, number, date, quote and citation as the source gives it, and each claim at the scope and certainty it was written with, because a proofread that changes what the text claims has edited its content.
  • Flag domain terminology you would change instead of replacing it, because a term that looks wrong to a general reader is often right for the intended one.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.