agentleFS
gigio1023gigio1023/gigio-skills/skills/additional/english-prompt-review/SKILL.md

english-prompt-review

Skill2 starsChanged 4 days agoMITView on GitHub

Use when a user wants an English task prompt reviewed or quietly coached while the requested work continues, especially for software engineering, AI research, and game-development work. Rewrites the complete request in clear, natural technical English, explains nuance and ambiguity in the user's native language (Korean by default), and keeps key expressions and contrasts in English. NOT for translating non-English text, proofreading prose word by word, or changing the user's task requirements.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

19 rules. About 0 are common, 17 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈17

    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. Distinctive:

    Read the whole English request before judging individual words.

    line 21
  2. Distinctive:

    Start or continue the requested work immediately.

    line 22
  3. Distinctive:

    If an ambiguous phrase could change task scope or technical meaning, do not select one interpretation.

    line 25
  4. Distinctive:

    Keep the user's requested outcome, constraints, exclusions, acceptance criteria, and authority unchanged.

    line 32
  5. Distinctive:

    When wording is materially ambiguous, do not hide the uncertainty inside a fluent rewrite.

    line 34
  6. Distinctive:

    Ask about ambiguity only where execution depends on it.

    line 36
  7. Distinctive:

    Treat the prompt as a small piece of working communication.

    line 40
  8. Distinctive:

    Prefer a plain, idiomatic request over either of these extremes:

    line 42
  9. Distinctive:

    Do not produce a line-by-line grammar lesson, a redline, a score, or a list of every small article and preposition change unless the user …

    line 47
  10. Distinctive:

    Keep the rewritten prompt, recommended expressions, contrast phrases, and reuse examples in English.

    line 51
  11. Distinctive:

    If delegation is unavailable, execute the task first and review the original prompt before writing the final response.

    line 65
  12. Distinctive:

    Read always-on-setup.md when the user wants this behavior applied automatically to ordinary English task requests.

    line 67
  13. Distinctive:

    Put the task result first.

    line 71
  14. Distinctive:

    Do not expose internal reasoning.

    line 99
  15. Distinctive:

    Before returning a rewrite, compare it with the original request:

    line 103
  16. Distinctive:

    Use evaluation-cases.md to test changes to this skill or to calibrate a harness integration.

    line 114
  17. Very specific:

    Do not correct fragmentary command prompts such as run the tests unless a rewrite would genuinely help the user learn a more useful …

    line 119
  18. Very specific:

    Do not turn a domain term into a prescribed implementation.

    line 120
  19. Distinctive:

    Do not let a fluent native-language explanation stand alone as proof of understanding.

    line 121

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

English Prompt Review

Help a user develop the habit of expressing technical work naturally in English without making the task harder to execute. Preserve the user's actual request; the coaching pass is a review, not a reinterpretation or a new requirement.

The native language below is the language the user's instructions name for explanations, else the language of their non-English messages; when neither shows it, use Korean, this pack's default, which the user's own instructions can change. The rewrite, recommended expressions, and contrasts stay in English whatever the native language is.

Quick Path

  1. Read the whole English request before judging individual words. Identify the outcome, relevant context, constraints, authority, and evidence of done.
  2. Start or continue the requested work immediately. Do not make the user wait for a language review.
  3. Re-express the request as one natural, direct prompt. Preserve the priority, scope, uncertainty, and level of urgency. Prefer the way an experienced practitioner would ask for the work, not a literal translation of the original syntax.
  4. Explain up to three high-value changes precisely in the native language, while keeping the recommended expressions, alternatives, and contrast examples in English.
  5. If an ambiguous phrase could change task scope or technical meaning, do not select one interpretation. Explain the distinction in the native language and show each possible meaning as an English clause or prompt variant.
  6. Return the coaching result after the task result. If the request is already concise and natural, the coaching section is the single line Natural and clear as written. instead of invented edits.

Coaching Contract

Preserve Meaning

Keep the user's requested outcome, constraints, exclusions, acceptance criteria, and authority unchanged. In particular, do not turn a request to inspect into a request to edit, soften a firm constraint, add a deadline, or substitute a different technical approach.

When wording is materially ambiguous, do not hide the uncertainty inside a fluent rewrite. Separate the shared meaning from the unresolved choice and give English variants for each plausible interpretation. Do not silently guess.

Ask about ambiguity only where execution depends on it. Continue independent authorized work; cosmetic coaching choices never block the main task.

Rewrite the Prompt, Not Every Token

Treat the prompt as a small piece of working communication. Use a clear subject and direct verb, put the desired outcome before implementation details, and group related constraints together. Omit filler that adds no decision value.

Prefer a plain, idiomatic request over either of these extremes:

  • literal translation from the user's native language;
  • inflated consultant or specification language when the user is simply asking an agent to do work.

Do not produce a line-by-line grammar lesson, a redline, a score, or a list of every small article and preposition change unless the user explicitly asks for that level of correction.

Split the Languages Deliberately

Keep the rewritten prompt, recommended expressions, contrast phrases, and reuse examples in English. Use the native language only to explain why a change matters, where an English expression's meaning begins and ends, or how an ambiguity would change the task.

Never place a full native-language translation directly under the rewrite. A native-language gloss alone can create recognition without command of the English expression. Tie each native-language explanation to the exact English phrase and, when a distinction is subtle, add an English contrast or example such as locate the failure versus identify the root cause.

Use Domain Language Carefully

Infer the domain from the request and repository context. Use a conventional technical term when it names a real distinction the original wording misses. Read terminology-guide.md when choosing a term for software engineering, AI/ML, or game development.

Keep exact non-sensitive technical names, identifiers, errors, commands, and code tokens. Never reproduce suspected secrets such as keys, tokens, passwords, private keys, connection strings, or signed URLs. Replace each with a stable typed placeholder such as <API_TOKEN_1> and reuse it consistently. When unsure, redact and note it. Keep accepted team vocabulary even when a synonym sounds more formal.

Keep the Main Work Moving

For background or delegated review, send only a copy sanitized with those placeholders. The reviewer returns a rewrite, optional terminology notes, and an ambiguity flag. The primary agent owns execution and the final response; never restore secrets in coaching output.

If delegation is unavailable, execute the task first and review the original prompt before writing the final response. Never claim that a review ran in the background when it did not.

Read always-on-setup.md when the user wants this behavior applied automatically to ordinary English task requests.

Output

Put the task result first. Then use this compact section only when coaching is enabled for the request or harness:

### English prompt coach

**Natural rewrite**

> <one complete prompt, or labeled complete English variants when the intended meaning is unresolved>

**한국어 가이드** *(도움이 될 때만)*

- `<preferred English expression>`: <의미 경계와 사용 이유를 설명하는 짧은 한국어>. `<English contrast or reuse example>`

**뜻 확인 필요** *(해석에 따라 작업이 달라질 때만)*

- `<ambiguous English phrase>`: <무엇이 확정되지 않았는지 한국어로 설명>
  - <의미 A>: `<English clause or prompt variant>`
  - <의미 B>: `<English clause or prompt variant>`

The template shows the Korean default. For another native language, write the two guidance headings, their conditions, and the explanations in that language.

For a prompt that needs no material rewrite, the section under its heading is the single line Natural and clear as written. in every mode that shows coaching. Do not praise mechanically, restate the task result, translate the full rewrite into the native language, or make the user compare near-identical versions. Offer one best rewrite unless a real ambiguity requires distinct English variants.

Authority and Privacy

  • A coaching request authorizes analysis and a suggested rewrite, not edits to the user's prompt, files, settings, or agent configuration.
  • The reviewer may use the current prompt and task context only. It must not send the prompt to an external service merely to judge wording.
  • Do not expose internal reasoning. Give the polished result and, when useful, a short reader-facing reason for a terminology recommendation.

Validation

Before returning a rewrite, compare it with the original request:

  • the same outcome, constraints, and boundaries remain;
  • verbs and nouns are natural for the inferred domain;
  • the prompt has one clear main request rather than a word-for-word gloss;
  • non-sensitive identifiers, commands, API names, and quoted content are unchanged, while secret values use stable typed placeholders;
  • key expressions and all alternatives remain in English;
  • native-language guidance explains semantic boundaries without translating the complete prompt or substituting for the English wording;
  • a material ambiguity is exposed as English alternatives rather than guessed;
  • any terminology note improves a real distinction rather than adding jargon.

Use evaluation-cases.md to test changes to this skill or to calibrate a harness integration.

Gotchas

  • Native-sounding English does not mean more formal, longer, or more idiomatic at the cost of the user's meaning.
  • Do not correct fragmentary command prompts such as run the tests unless a rewrite would genuinely help the user learn a more useful expression.
  • Do not turn a domain term into a prescribed implementation. For example, root cause describes a diagnosis goal; it does not authorize code changes.
  • Do not let a fluent native-language explanation stand alone as proof of understanding; anchor it to an English expression, contrast, or reuse example.

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.