agentleFS
hoshiori-devhoshiori-dev/harness.md/.agents/skills/human-writing/SKILL.md

human-writing

Skill0 starsChanged 4 days agoApache-2.0View on GitHub

Drafts, edits, and reviews text meant for human readers — articles, essays, announcements, documentation prose, newsletters — in English, Chinese, or Japanese. Use when writing or improving any piece a person will read; when a draft "sounds like AI", should be "more human", or needs polish without losing the author's voice; when reviewing whether a text lands with its intended audience; or before publishing prose produced as a side effect of other work. Also applies as the shared baseline when a genre-specific writing skill (academic, blog, copy) is active — the specialized skill's rules win on conflict. Not for code, commit messages, or agent-facing documents such as AGENTS.md, skill instructions, or knowledge-base entries.

46 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use leader, storage-analyzer and 4 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • baizhine999Skill

    leader

    AI 扮演你的 Leader,用大厂管理思维审视你的方案、想法和工作,push 你成长。触发词:/leader, leader模式, 大厂模式, 帮我看看这个, 我有个方案, 我最近有点迷茫, 我不知道该怎么做。AI 是施压者,你是被 push 的那个人。

    Also in 3 projects that use this

  • xiaolinVVSkill

    storage-analyzer

    macOS / Windows 只读存储分析助手(自动识别系统)。扫描整机磁盘占用,找出 占空间大户,把每一项分成 🟢可自动清理 / 🟡需人工判断 / 🔴谨慎清理 三级并给出 可执行处置方案,生成排版精美、可折叠、命令可一键复制的交互式 HTML 报告,并可 起本地服务在网页上一键删除(移废纸篓/直接删)。扫描全程只读。务必在以下场景 使用:用户说"存储分析""磁盘满了""C盘/硬盘满了""空间不够""清理空间" "清理磁盘""占空间""哪些东西占地方""帮我看看存储""看一下电脑存储/空间" "存储空间""电脑空间不够""内存满了/不够/不足""看下内存/存储"(中文口语里 "内存"常指存储空间)"storage analysis""disk cleanup""清缓存""磁盘清理"; 或用户抱怨电脑没空间、想知道什么东西吃硬盘、想要清理建议时。注意:若用户明确 指运行内存/RAM(如"哪个进程吃内存""内存占用高"想看活动监视器),那是 RAM 不是存储,不属于本 skill。

    Also in 3 projects that use this

  • Wyx1105170322Skill

    neat-freak

    End-of-session knowledge cleanup with OCD-level rigor — reconciles project docs (CLAUDE.md, README.md, docs/) and agent memory against the code so nothing rots. 会话结束后对项目文档和记忆进行洁癖级审查与同步。MUST trigger when the user says: "sync up", "tidy up docs", "update memory", "clean up docs", "/sync", "/neat", "同步一下", "整理文档", "整理一下", "更新记忆", "梳理一下", "收尾", "这个阶段做完了", "新人能直接上手", or any phrase suggesting a dev milestone where knowledge needs reconciliation. Also trigger when the user reports stale docs, conflicting memories, or wants a clean handoff to teammates or other agents. Bare "整理" / "tidy" with prior dev context counts — do not under-trigger. Cross-platform: works on Claude Code, OpenAI Codex, OpenCode, and OpenClaw.

    Also in 3 projects that use this

  • NVIDIASkill

    deep-research

    Queue deep, multi-step research and analysis tasks to the DeepAgents worker. Supports execution depth presets, request-specific rubrics, SubAgent delegation, and RubricMiddleware cross-validation.

    Also in 3 projects that use this

  • VercelSkill

    find-skills

    Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.

    Also in 3 projects that use this

  • SentrySkill

    skill-creator

    Create new agent skills following the Agent Skills specification. Use when asked to "create a skill", "add a new skill", "write a skill", "make a skill", "build a skill", or scaffold a new skill with SKILL.md. Guides through requirements, writing, registration, and verification.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

12 rules. 0 are common, 12 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 29 edited versions we read keep these.

  • 12

    Distinctive

    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
  1. Distinctive:

    Before writing or editing a single sentence, pin down four things — from the user's request, the existing text, or by asking:

    line 32
  2. Distinctive:

    Keep the author's explicit stance.

    line 58
  3. Distinctive:

    Add no defensive qualifiers ("it's important to note", "of course, this depends") unless the uncertainty is real and the author would state …

    line 60
  4. Distinctive:

    Commit to claims. An author who believes something writes it plainly.

    line 62
  5. Distinctive:

    Treat these as don'ts while drafting and as a checklist while reviewing:

    line 68
  6. Distinctive:

    Every external source — book, paper, case, statistic — must exist and must meet the credibility bar the piece's form demands.

    line 94
  7. Distinctive:

    After drafting or editing, review from outside the author's seat with three checks:

    line 102
  8. Distinctive:

    Language and wording review: if the environment can run subagents with a separate, clean context, send the piece with …

    line 105
  9. Distinctive:

    Reader review: if clean-context subagents are available, send the piece with assets/reader-review-prompt.md filled in.

    line 112
  10. Distinctive:

    AI-writing detection: if clean-context subagents are available, send the text (with no authorship context) using …

    line 117
  11. Distinctive:

    When the environment can run subagents in parallel, dispatch all three checks at once and collect their findings together before changing …

    line 123
  12. Distinctive:

    If the input text is already natural and idiomatic, say which parts are good and change nothing — do not edit to justify the invocation.

    line 131

“Common” is measured: kept by at least 6 in 10 of the 29 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

Human Writing

Text for human readers fails differently from text for machines: the most resented AI writing is not wrong, it is authorless — smooth, correct, and empty. Everything below exists to keep a specific author present in the text.

Scope and precedence

This skill covers only text whose reader is a person. Code, commit messages, and agent-facing documents (AGENTS.md, skill files, knowledge bases) follow their own conventions — do not apply this skill to them. When a genre- or project-specific writing skill is also loaded, apply both with the specialized skill leading: its requirements override this baseline wherever they conflict.

1. Define the piece before touching it

Before writing or editing a single sentence, pin down four things — from the user's request, the existing text, or by asking:

  • Type: what kind of piece this is, and the conventions its venue implies.
  • Author: who is speaking — their identity, stance, and how they want to appear.
  • Reader: who receives it — background, expectations, what they already believe.
  • Effect: what the author wants to change in the reader — what they should feel, think, or do afterward.

From these four, derive the piece's center (emotion, opinion, or logic), its voice and register, and an explicit do/don't list. Human authors write from a concrete identity and stance; erasing that presence is the core failure of AI writing.

2. Plan with budgets, then draft

Outline before drafting, top-down: order sections by importance to the intended effect, and give each a length budget proportional to its value. A piece where every point gets equal space says that nothing in it matters. Only start prose once the structure and budgets stand.

3. Write as the author, not as an assistant

Reinforcement learning on user feedback rewards defensive, ingratiating, verbose, convergent text. Step outside that pull by cycling roles: create as the author, revise as an editor, judge as the reader. Concretely:

  • Keep the author's explicit stance. Never rewrite it into mechanical neutrality or bury it under "balanced" both-sides padding the author did not ask for.
  • Add no defensive qualifiers ("it's important to note", "of course, this depends") unless the uncertainty is real and the author would state it.
  • Commit to claims. An author who believes something writes it plainly.

4. Failure modes to avoid

A digest of recognizable AI-writing patterns (fuller catalog: Wikipedia's Signs of AI writing). Treat these as don'ts while drafting and as a checklist while reviewing:

  • Content: inflated significance ("stands as a testament", "pivotal role in the evolving landscape"); superficial analysis bolted on with "-ing" clauses ("..., highlighting the importance of..."); promotional gloss on neutral subjects; opinions attributed to no one ("experts say", "widely regarded"); formulaic closing sections about challenges and future prospects.
  • Language: high density of AI-flavored vocabulary (delve, underscore, pivotal, robust, landscape, tapestry, meticulous, showcase); avoiding "is/are" in favor of "serves as / represents / functions as"; negative parallelisms ("not just X, but Y") and rule-of-three chains as default rhythm; rotating synonyms to dodge honest repetition.
  • Style: mechanical bold emphasis; headers-with-colons bullet lists standing in for paragraphs; em-dash overuse; emoji as structure; every section the same length and shape.
  • Citations: sources that do not exist, do not say what is claimed, or carry broken links and tracking parameters.

These patterns surface differently per language. Read references/english.md, references/chinese.md, or references/japanese.md — matching the language the piece is written in — before drafting or reviewing in that language.

5. Citations

Every external source — book, paper, case, statistic — must exist and must meet the credibility bar the piece's form demands. Verify before citing; if a source cannot be verified, drop the claim rather than keep an unsourced or invented reference. LLMs fabricate plausible citations to keep prose flowing; treat every reference you did not verify as suspect.

6. Review loop

After drafting or editing, review from outside the author's seat with three checks:

  • Language and wording review: if the environment can run subagents with a separate, clean context, send the piece with assets/language-review-prompt.md filled in per its instructions; otherwise review it yourself against the same questions. Is the text fluent in this language, is the grammar right, and is every word one the target readers actually use — common words over obscure or literary picks, unless the register truly demands otherwise?
  • Reader review: if clean-context subagents are available, send the piece with assets/reader-review-prompt.md filled in; otherwise re-read the piece yourself in the reader role, against the same questions. Did the piece achieve the intended effect? Does the structure carry it?
  • AI-writing detection: if clean-context subagents are available, send the text (with no authorship context) using assets/ai-detection-prompt.md, filled in from the failure-mode digest and the matching language reference per the template's own instructions; otherwise skip this pass — the failure-mode digest above already served as your self-check.

When the environment can run subagents in parallel, dispatch all three checks at once and collect their findings together before changing anything. When running sequentially — no parallel support, or the self-review fallbacks — do the language and wording check first, in the order listed. Fix what the reviews surface, then re-review. Stop when a pass reports no substantive findings.

7. Restraint

If the input text is already natural and idiomatic, say which parts are good and change nothing — do not edit to justify the invocation. Before final output, ask: did I change any fluent sentence only to leave a mark? If so, revert it. Fewer, correct edits beat many defensible ones.

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.