agentleFS
MrD-LtMrD-Lt/codex-final-review/SKILL.md

final-review

Skill2 starsChanged 5 days agoMITView on GitHub

Use only after task work and a final draft are complete, immediately before delivery, to review the wording of the final response or artifact prose.

6 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions6 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 6

    1. DanMcInerney/architect-loop★ 626same name
    2. aromanarguello/roman-skills★ 18same name
    3. normalled/apijack★ 7same name
    4. limpidwei/final-exam-review★ 5same name
    5. jsuvic/agent-profile★ 3same name
    6. andeeplus/aplus★ 2same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

Show its rules and text

We split this skill into the rules it gives an agent

27 rules. 0 are common, 20 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 20

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Treat this as a final wording pass over an already completed draft.

    line 13
  2. Distinctive:

    Use evidence already gathered to check the draft.

    line 16
  3. Distinctive:

    Use the language the user explicitly requests.

    line 21
  4. Distinctive:

    If no output language is specified, reply in the language used for the user's instructions and questions in the current request.

    line 22
  5. Distinctive:

    When the request is mixed or language-neutral, preserve the established conversation language.

    line 23
  6. Distinctive:

    Do not treat code, identifiers, file names, quotations, or source material as a request to switch languages.

    line 24
  7. Very specific:

    Read references/custom-language-rules.md completely before reviewing.

    line 25
  8. Distinctive:

    Treat the user's current explicit instruction and the source meaning as authoritative when they conflict with a persistent custom language …

    line 26
  9. Very specific:

    When the selected output language is English, read references/humanizer.md completely and apply its Embedded mode together with this skill.

    line 27
  10. Very specific:

    If references/settings.yaml exists, read it completely before choosing conditional references.

    line 32
  11. Very specific:

    Treat the experimental code review as enabled only when code_review.enabled is the YAML boolean true.

    line 33
  12. Very specific:

    When code review is enabled and the output concerns code, configuration, tests, builds, or implementation reporting, read …

    line 34
  13. Very specific:

    When code review is disabled, skip references/code-review.md and continue applying all shared factual, scope, language, and delivery rules.

    line 35
  14. Very specific:

    If code review is enabled but references/code-review.md is absent, continue with the shared review and do not claim that a code-specific …

    line 36
  15. Distinctive:

    When accuracy could materially affect the answer and verification is reasonably available, verify names, numbers, dates, versions, …

    line 45
  16. Distinctive:

    Keep content that helps answer the question or complete the task.

    line 49
  17. Distinctive:

    Remove unsupported or irrelevant background, atmosphere, motives, causes, and narrative completion.

    line 50
  18. Distinctive:

    Remove unrequested expansion, generic advice, and decorative completeness.

    line 51
  19. Distinctive:

    Remove templated pleasantries, unsupported praise, repeated greetings, process announcements, repeated summaries, and routine closing …

    line 65
  20. Distinctive:

    Keep simple answers short and expand complex answers only as needed for understanding.

    line 66
  21. Distinctive:

    Use headings, lists, bold text, and summaries when they improve reading.

    line 67
  22. Distinctive:

    Prefer concrete, plain, natural wording.

    line 68
  23. Distinctive:

    When the overall view is favorable and some details need qualification, state the overall view first and connect the qualifications …

    line 72
  24. Distinctive:

    Use specific ordinary wording in place of editorial shorthand, administrative language, and unnecessary technical labels.

    line 73
  25. Distinctive:

    Use first- and second-person pronouns when they clarify responsibility, preference, source, or interaction.

    line 75
  26. Distinctive:

    Check whether the response answers the current question and avoids agreement without evidence.

    line 80
  27. Distinctive:

    Remove vague attribution, promotional tone, inflated significance, slogans, forced groups of three, translation-like prose, repeated …

    line 83

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “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

Final Review

Apply this skill only after the task work and final draft are complete, immediately before delivering the final response or artifact prose. Review the wording silently, revise it in place, and return the revised result without an audit report.

Timing and scope

  • Complete research, reasoning, calculations, tool use, implementation, and required verification using the task's normal instructions before invoking this skill. Do not read or apply its language references during those stages or for progress updates.
  • Treat this as a final wording pass over an already completed draft. Preserve substantive conclusions, supporting explanations, necessary distinctions, uncertainty, recommendations, and requested detail. Language preferences must not determine what to investigate or which conclusion to reach.
  • Judge wording in the context of the user's actual question. Establish which judgments and qualifications are necessary using the relevance review below, then preserve their substance while improving their phrasing.
  • Review prose in final responses, documents, comments, and completion reports. Do not change executable code, configuration behavior, calculations, tests, or tool decisions as part of this wording pass.
  • Use evidence already gathered to check the draft. If review reveals a substantive error or a need for new verification, return to the task workflow, resolve it, and then review the revised final draft. Do not substitute a smoother statement for unresolved work.
  • These instructions define intended invocation timing, not a technical isolation mechanism. Once read, they remain in the conversation context; do not apply their stylistic rules to later task reasoning.

Language routing

  • Use the language the user explicitly requests.
  • If no output language is specified, reply in the language used for the user's instructions and questions in the current request.
  • When the request is mixed or language-neutral, preserve the established conversation language.
  • Do not treat code, identifiers, file names, quotations, or source material as a request to switch languages.
  • Read references/custom-language-rules.md completely before reviewing. Apply only the section that matches the requested output language.
  • Treat the user's current explicit instruction and the source meaning as authoritative when they conflict with a persistent custom language preference.
  • When the selected output language is English, read references/humanizer.md completely and apply its Embedded mode together with this skill. This skill's factual grounding, relevance and scope control, and matching custom language rules take precedence over conflicting Humanizer surface-style rules.
  • For other output languages, apply the shared factual, relevance, scope, and voice rules together with any matching custom language section. Leave the English Humanizer reference inactive.

Review configuration

  • If references/settings.yaml exists, read it completely before choosing conditional references.
  • Treat the experimental code review as enabled only when code_review.enabled is the YAML boolean true. If the file is absent, malformed, or uses another value, treat code review as disabled.
  • When code review is enabled and the output concerns code, configuration, tests, builds, or implementation reporting, read references/code-review.md completely before reviewing.
  • When code review is disabled, skip references/code-review.md and continue applying all shared factual, scope, language, and delivery rules.
  • If code review is enabled but references/code-review.md is absent, continue with the shared review and do not claim that a code-specific review was performed.

Review order

Facts and status

  • Ground every fact, action, background detail, state, motive, cause, and result in the user input, inspected files, tool output, or a reliable source.
  • Mark confirmed information, inference, recommendation, and uncertainty accurately.
  • Report file reads, commands, tests, checks, and completion status from actual evidence.
  • When accuracy could materially affect the answer and verification is reasonably available, verify names, numbers, dates, versions, quotations, links, and citations. When verification is unavailable, state the source and uncertainty instead of implying confirmation.

Relevance and scope

  • Keep content that helps answer the question or complete the task.
  • Remove unsupported or irrelevant background, atmosphere, motives, causes, and narrative completion.
  • Remove unrequested expansion, generic advice, and decorative completeness.

For qualifications, warnings, and corrective explanations, decide what belongs before polishing how it sounds. Review each clause in the context of the whole answer:

  1. Identify its job. Tie the clause to a question or belief the user actually expressed, a condition governing the answer's validity or scope, or an evidenced constraint affecting the requested action. A fact being true or a misunderstanding being common does not by itself establish relevance.
  2. Test its contribution. Temporarily remove the clause. Identify the specific answer, applicability condition, confidence level, or next action that would become materially wrong or incomplete. Judge what the remaining answer already communicates; a boundary stated once does not also need a warning against crossing it.
  3. Choose the treatment. Keep a clause that supplies necessary information. When necessary information is wrapped in an imagined mistaken inference, express the information directly and remove the corrective framing. Delete a clause whose only contribution is rejecting an unexpressed belief or repeating an already clear limit; do not replace it with a softer warning.
  4. Check the result. Confirm that the revised answer still addresses any explicit misconception or comparison and retains concrete prerequisites, consequential risks, and evidence limits. A material constraint can be necessary even when the user did not ask about it; establish that need from the actual task and available evidence, not a hypothetical reader's possible mistake.

Apply this check to positive-sounding reminders as well as negative sentences. Preserve substantive meaning in requested quotations, translations, and faithful rewrites; assess relevance when editing the assistant's own explanatory additions. Keep these decisions internal.

Natural delivery

  • Lead with the answer.
  • Remove templated pleasantries, unsupported praise, repeated greetings, process announcements, repeated summaries, and routine closing questions. Preserve natural politeness and emotional acknowledgment when the genre, relationship, or user's situation calls for them.
  • Keep simple answers short and expand complex answers only as needed for understanding.
  • Use headings, lists, bold text, and summaries when they improve reading.
  • Prefer concrete, plain, natural wording.
  • Match the user's wording, rhythm, politeness, and expertise when a sample is available.
  • In ordinary discussion, use a collaborative, conversational tone. Remove adjudicative openings and report-like phrasing.
  • Begin with a complete sentence. Use verdict or summary labels only when the user requests a formal review, decision memo, or summary.
  • When the overall view is favorable and some details need qualification, state the overall view first and connect the qualifications naturally.
  • Use specific ordinary wording in place of editorial shorthand, administrative language, and unnecessary technical labels. Describe the needed adjustment in concrete terms.
  • Preserve qualifiers and transitions that make the prose natural. Concision should still sound like ordinary conversation rather than compressed report prose.
  • Use first- and second-person pronouns when they clarify responsibility, preference, source, or interaction. In neutral recommendations, make the relevant object, condition, or action the center of the sentence.
  • For practical instructions, state the timing or condition, object, action, method, and purpose when each is needed. Split a dense sequence into separate sentences or steps.

Conversation and document review

  • Check whether the response answers the current question and avoids agreement without evidence.
  • Preserve the source meaning, responsible party, commitment level, uncertainty, terminology, names, numbers, dates, and citations.
  • Retain the user's voice and document genre.
  • Remove vague attribution, promotional tone, inflated significance, slogans, forced groups of three, translation-like prose, repeated conclusions, and excessive sectioning.
  • Describe the current state in ordinary documentation. Use change-centered narration for changelogs, release notes, and migration guides when the format calls for it.

Delivery

Return the revised final content. Keep the review process, checklist, draft, and audit commentary hidden. State unresolved uncertainty when the available evidence cannot support a stronger claim.

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.

Reviews can't be read right now.

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.