agentleFS
drosamdrosam/agents/skills/writing-for-humans/SKILL.md

writing-for-humans

Skill1 starsChanged 2 days agoMITView on GitHub

Writes and edits text that other people will read in collaboration tools (GitHub, GitLab, Linear, Jira, Slack, Teams and similar) so it is short, direct and sounds like a person, without dropping facts. Use when drafting, rewriting or posting PR/MR review comments and replies, PR/MR descriptions, issue, ticket or work-item bodies and comments, status updates, or chat messages, or when asked to make such text shorter, less wordy, less robotic or more human. Not for replies to the user in the current conversation, code comments, documentation files or commit messages.

9 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use retro.

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.
  • AWSSkill

    retro

    Conduct a retrospective on a coding session.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 8 edited versions we read keep these.

  • 10

    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:

    Write the way a busy, friendly senior engineer types: point first, plain words, nothing the reader already knows, nothing they need left …

    line 8
  2. Distinctive:

    Write full natural sentences in the draft even when the conversation uses a compressed or terse style.

    line 13
  3. Distinctive:

    Cut: Remove every tell in the cut list below.

    line 23
  4. Distinctive:

    Final check: Run the pre-publish checklist.

    line 25
  5. Distinctive:

    Use the target's syntax. When unsure, plain text with code spans.

    line 72
  6. Distinctive:

    No headings in comments or chat.

    line 75
  7. Distinctive:

    Every fact, number, name, link and error from the source survived.

    line 81
  8. Distinctive:

    Every sentence adds something the reader did not have.

    line 82
  9. Distinctive:

    No cut-list tell remains.

    line 83
  10. Distinctive:

    No secrets, tokens, credentials, customer data or private URLs the audience should not see.

    line 85

“Common” is measured: kept by at least 6 in 10 of the 8 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

Writing for Humans

Write the way a busy, friendly senior engineer types: point first, plain words, nothing the reader already knows, nothing they need left out.

Scope

  • Apply only to the text being drafted for publication. The conversation with the user keeps its own style; switch back after the draft.
  • Write full natural sentences in the draft even when the conversation uses a compressed or terse style. Fragments and shorthand read as robotic in public.
  • This skill shapes wording only. It grants no permission to post; follow the caller's or user's approval rules.
  • A repo, team or tracker template wins over the defaults below. Fill the template in this voice.

Workflow

  1. Reader and action. Name who reads it and what they must do: fix, answer, decide, review, or nothing. Note what they already know from the thread, ticket or PR.
  2. Point first. The first sentence carries the ask, answer, decision or problem. No run-up.
  3. Keep the facts. List what the reader needs to act and keep all of it: error text, values, versions, paths, links, repro steps, impact, who is affected, deadlines. Shorten wording, never data.
  4. Add only what helps. One reason when the ask is not obvious. A short example, snippet or repro when it makes the problem faster to see. Bullets when there are 3+ parallel items (steps, criteria, options, affected cases).
  5. Cut. Remove every tell in the cut list below.
  6. Check length. Compare with the budget. Over budget: cut wording first; if every remaining line is a needed fact, keep it. Never compress into fragments to fit.
  7. Final check. Run the pre-publish checklist.

Length budgets

Defaults, not caps. Needed facts beat the budget; filler never does.

TextDefault
Inline review comment1-3 sentences, plus a snippet or suggestion block if it helps
Reply in a thread1-2 sentences; often one
PR/MR description1-3 sentence summary of what and why; short bullets for notable changes, testing and risks only when present
Bug reportTitle + 1-2 sentence summary + steps, expected, actual, environment
Feature request / ticketTitle + problem and outcome in 1-3 sentences; acceptance criteria as bullets
Ticket comment / status update1-3 sentences: state, next step, blocker or ask
Chat message1-3 sentences in one message; thread long detail

Ceiling: it fits on one screen, unless the content is itself a log, table or list the reader asked for.

Voice

  • Plain words and contractions (it's, don't, I'd). Talk about the code or the change, not the person.
  • State what you know; ask when you are unsure. One hedge at most per claim.
  • Mark weight when it matters: nit:, optional:, non-blocking:, blocking:. Use the team's label convention if it has one; do not introduce a new one.
  • Match the thread's language, register and formatting. When recent comments from the same team are available, mirror their length and tone.
  • Thanks or praise only when genuine and specific, once, briefly.
  • Disagree calmly: the fact, the consequence, the proposal. No defensiveness, no apology chains.

Cut list

TellLooks likeDo instead
Chat wrapper"Great question!", "Happy to help", "I hope this helps", "Let me know if you have any questions"Delete
Run-up"I wanted to flag", "Here's the thing", "It's worth noting that", "Just a heads up"State the point
Recap the reader hasReply restates the problem and diagnosis before the answerAnswer first; keep only the fact they lack
Work narration"I've carefully reviewed", "After analyzing the code", "I went ahead and"Delete; give the result
Fake contrast"This isn't just X, it's Y", "Not X, but Y"Say Y
Inflated wordscrucial, robust, comprehensive, seamless, leverage, ensure, enhance, delve, significant, keyPlain word or delete
Stacked hedges"might potentially", "could possibly", "it seems like it may"One hedge or none
Over-politeness"Would you mind possibly considering""Could you...?"
DecorationHeadings or bold labels in short text, emojis, a bullet list for two itemsProse
Forced threes"faster, safer, and more maintainable" when one is trueKeep only the real items
Dashes everywhereem or en dashes joining clausesComma, period or parentheses, unless the user's own writing uses them
Summary closer"Overall, this change improves...", "In summary"Delete
Robot shorthand"Null deref L42. Fix: early return."Full short sentences

Formatting

  • Use the target's syntax. When unsure, plain text with code spans.
  • Code spans for identifiers, paths and values; fenced blocks for multi-line code or logs; trim logs to the relevant lines.
  • Links instead of pasted long URLs. Refer to issues, commits and PRs with the platform's link form.
  • No headings in comments or chat. In descriptions, use headings only when a template has them or the body has several distinct parts.

Pre-publish checklist

  • First sentence carries the point.
  • Reader knows what to do, or that nothing is needed.
  • Every fact, number, name, link and error from the source survived.
  • Every sentence adds something the reader did not have.
  • No cut-list tell remains.
  • Within budget, or every extra line is a needed fact.
  • No secrets, tokens, credentials, customer data or private URLs the audience should not see. Redact log and screenshot content.

Examples

Review comment:

Bad: Great work on this PR! I noticed a potential issue that might be worth considering. It seems that fetchUser could potentially return null in certain edge cases, which may lead to unexpected behavior. It would be beneficial to add a null check to ensure robustness. Let me know what you think!

Good: fetchUser returns null when the session has expired, and line 42 reads .id from it. Could we return early here?

Reply after fixing:

Bad: Thank you so much for catching this! You're absolutely right. I've gone ahead and updated the code to properly handle this case by adding validation.

Good: Good catch, fixed in a1b2c3d.

More cases (PR description, bug report, ticket, status update, chat, disagreement, secrets, over-shortened text): read references/examples.md when drafting one of those types or when a draft still sounds robotic after the cut list.

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.