willey-labs/agent-skills · Archived

writing-standards

Standards for writing, editing, and reviewing DOCUMENTS — READMEs, specs, rules, skills, design docs, guides, and any prose deliverable produced from a source (existing code, or a prior discussion). Two rule sets apply on every document task: (1) source-to-deliverable — code→doc describes what the system DOES with no code or code-identifiers; discussion→rule states the general principle, not the specific example just discussed; (2) anti-slop — cut hedging, hype words, throat-clearing, cheerlead…

First seen Jun 28, 2026

Installation

$ npx skills add willey-labs/agent-skills --skill writing-standards

Stronger alternatives

This repository is archived — consider an actively maintained alternative.

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from willey-labs/agent-skills.

npx skills add willey-labs/agent-skills

Browse all from willey-labs/agent-skills

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Repository health

Stars 1
License MIT
Default branch main
Open issues 0
Status Archived

Skill metadata

Parsed from SKILL.md frontmatter.

Version0.1.0
LicenseMIT
Declared agents claude-code
More metadata
author
willey-labs
version
0.1.0

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 3,952 B
  • docs SUMMARY.md 1,001 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 9 installs

SKILL.md

Writing Standards

Every document you write, edit, or review must comply with these rules. A document is any prose deliverable: a README, a spec, a rule, a skill, a design doc, a guide, a review, release notes. This skill does not govern normal conversation — only the deliverable.

The rules come in two sets, both applied on every document task:

  • references/common/source-to-deliverable.md — how to turn a source (code you read, or a

discussion you had) into a deliverable without dragging the source into the output.

  • references/common/anti-slop.md — the patterns that make a document read as machine-padded

filler, each with a rule code and a fix.

All paths in this document are relative to this SKILL.md file, so they resolve wherever the skill is installed.


Step 0 — Bootstrap the reminder hooks (once per session, only if needed)

The hooks inject a short reminder of these rules at session start and on every prompt, so the rules don't get buried as a conversation grows. Run the fast read-only check first (single absolute-path command, no cd/&&):

python3 <skill-dir>/bootstrap.py --verify

Exit 0 → hooks are wired; go to Step 1. Non-zero → wire them:

python3 <skill-dir>/bootstrap.py

It auto-detects scope (this project's .claude/settings.json vs global ~/.claude/settings.json) from where the skill is installed, and merges two entries — SessionStart and UserPromptSubmit — without disturbing hooks you already have. Tell the user to restart the session so the hooks activate; until then the reminder isn't injected, but this SKILL.md is still in force for any document you write this session.


Step 1 — Pick the document type (optional refinement)

references/common/ applies to every document. If a per-type guide exists under references/<doc-type>/ (README, spec, review, …), read it too — it specializes the common rules for that kind of document. If none matches, the common rules are the whole standard.


Step 2 — Write, edit, or review against the rules

Read both common references, then produce the document. The two failures this skill exists to stop:

  1. Echoing the source. You read code (or had a discussion) to understand something, then paste

that source — the code, the function names, the one example you just discussed — into the deliverable. The source is what you learned from; it does not appear in the output. See source-to-deliverable.md.

  1. Slop. Hedging, hype, throat-clearing, cheerleading, reflexive headers/bold/lists, and

sentences that sound like content but carry none. See anti-slop.md.

When reviewing a document, report violations by rule code with the offending line and the fix — the same way coding-standards reports a code review.