documentation-writer
Writes and revises human-facing documentation — READMEs, contributor and setup guides, reference pages, architecture notes, CLAUDE.md, and changelog entries.
Skill by dougborg
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/dougborg-documentation-writer
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: documents, writing, productivity, agent-building
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/documentation-writer
Tools it pre-approves 5
From allowed-tools in SKILL.md: the agent may use these without asking you each time.
Read,Write,Edit,Glob,Grep
What's inside
- Skill name
documentation-writer- Description
- Writes and revises human-facing documentation — READMEs, contributor and setup guides, reference pages, architecture notes, CLAUDE.md, and changelog entries. Covers picking the right document type for the question being asked, structuring for skimming, writing examples that actually run, and keeping docs from going stale. Use when creating or rewriting docs, when a README has drifted from the code, or when deciding where a piece of information belongs. For SKILL.md and agent files, use /skill-writer instead.
- License
- Apache-2.0
- Allowed tools
- Read,, Write,, Edit,, Glob,, Grep
- Instructions
- 110 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# /documentation-writer — Human-Facing Documentation
Documentation fails in two directions: it does not answer the question the
reader arrived with, or it answers it somewhere the reader never looks. Most of
the work is deciding what kind of document you are writing and who is reading
it, before writing a line.
## Identify the document before writing it
Readers arrive with one of four questions. Mixing the answers into one page is
the most common documentation failure — a tutorial interrupted by API tables
serves neither reader.
| Reader's question | Document type | Shape |
| --- | --- | --- |
| "Get me to a first success" | Tutorial / quickstart | Ordered steps, one path, no options |
| "How do I accomplish X?" | How-to guide | Task-titled, assumes basics, may branch |
| "What are the exact parameters?" | Reference | Complete, structured, alphabetical or grouped |
| "Why is it built this way?" | Explanation / ADR | Prose, trade-offs, alternatives rejected |
If a page is trying to be two of these, split it and cross-link.
## Write for scanning
Readers scan before they read. Every structural choice should reward a reader
who only looks at headings and the first line of each section.
- **Headings state the answer, not the topic.** "Run migrations before starting
the server" beats "Migrations."
- **Front-load.** The first sentence of a section carries its conclusion; the
rest supports it.
- **One idea per paragraph**, and prefer a table or list when the content is
genuinely parallel — but not when it is prose wearing a list costume.
- **Link out instead of inlining depth.** Name what is on the other side of the
link so the reader can judge whether to follow it.
Length follows the content. A README for a small tool may be thirty lines; a
reference page may be five hundred. There is no target word count — the test is
whether a reader can find their answer, not how long the page is.
## Examples must run
Copy-pasteable examples are the highest-value content in most documentation,
and wrong ones are the highest-cost.
- Show real commands and real output, not placeholders the reader must decode.
- Verify examples against the current code before publishing. If you cannot run
it, say what version it was checked against.
- Prefer a complete small example over a fragment with elisions.
- Include the failure case when it is common — the error message a reader will
paste into a search box is worth documenting verbatim.
## Put information where it will be found
The right location beats the right words. Before writing, check whether the
information already lives somewhere and is simply stale.
- **README** — what this is, who it is for, how to run it, where to go next.
Not a manual.
- **Code comments** — why this line is surprising. Never what the line does.
- **`CLAUDE.md`** — the conventions an agent must hold to work in this repo.
Keep it short; it loads on every session.
- **Reference docs next to the code they describe**, so a change and its
documentation land in the same diff.
- **Changelog / release notes** — what changed for a consumer, in their terms,
not commit subjects.
Duplicating content across locations guarantees divergence. Pick one home and
link to it.
## Keep it from rotting
Stale documentation is worse than none, because readers trust it.
- Avoid time-relative phrasing — "currently", "new", "recently", "for now",
"the upcoming release". Write what is true, or date the claim.
- Do not document unreleased or intended behavior as if it exists.
- When you change behavior, grep for the old behavior in the docs in the same
change.
- Prefer generated reference material (CLI `--help`, schema dumps) over
hand-maintained tables that will drift.
- If you find documentation that is wrong, fix it or delete it. Leaving it with
a caveat is the worst of the three options.
## Voice
- Second person and active voice: "run the migration", not "the migration
should be run."
- Present tense for behavior: "the hook blocks the commit."
- Define a term once, then use exactly that term — synonyms read as new
concepts.
- Say what a thing does before saying what it is called.
- Cut hedges ("simply", "just", "of course", "obviously"). They add nothing and
make a stuck reader feel worse.
## Before you ship
- Does the first paragraph tell a stranger whether this page is for them?
- Can a reader who skims only the headings get the gist?
- Does every command and code block run as written?
- Is anything here duplicated from another document?
- Would this still be correct in six months, or does it need a date?
## Related
- `/skill-writer` — SKILL.md and agent files, which have different constraints
(they load into a model's context and carry frontmatter contracts)
- `CLAUDE.md` — this project's own conventionsFrom dougborg/harness-kit, Apache-2.0.
More from dougborg/harness-kit
- 0.565score
adhd-mode
dougborg
Shapes every response for a reader with ADHD: the next action first, numbered steps, progress restated each turn, tangents held back, concrete time estimates…
SkillAFreshness: AClaude Code +9 - 0.565score
agent-standup
dougborg
Reconciles and coordinates shared work across multiple coding agents and human operators.
SkillAFreshness: AClaude Code +9 - 0.565score
code-reviewer
dougborg
Reviews a diff across six dimensions — correctness, design, readability, performance, testing, security — plus the repo's documented standards, code smells…
SkillAFreshness: AClaude Code +9 - 0.565score
codebase-design
dougborg
Shared vocabulary and principles for designing deep modules: a lot of behaviour behind a small interface, at a clean seam, testable through that interface.
SkillAFreshness: AClaude Code +9 - 0.565score
commit
dougborg
Creates conventional commits with quality gates — stages intentionally, checks for uv.lock drift, runs the project's verification command, and writes the…
SkillAFreshness: AClaude Code +9 - 0.565score
diagnosing-bugs
dougborg
A disciplined diagnosis loop for hard bugs and performance regressions: build a tight feedback loop that goes red on this bug, minimise, hypothesise…
SkillAFreshness: AClaude Code +9 - 0.565score
domain-modeling
dougborg
Builds and sharpens a project's domain language as decisions are made: challenges fuzzy or conflicting terms, stress-tests concepts with edge-case scenarios…
SkillAFreshness: AClaude Code +9 - 0.565score
grilling
dougborg
Interviews the user relentlessly about a plan, design, decision, or idea until every branch of the decision tree is resolved and nothing is silently assumed.
SkillAFreshness: AClaude Code +9 - 0.565score
harness-issue
dougborg
Files a bug report, idea, or proposed change against the upstream harness repo resolved from the project's config, from any project that uses harness-kit.
SkillAFreshness: AClaude Code +9 - 0.561score
budget
dougborg
Check the Claude usage/token budget before dispatching subagents or large fan-outs.
SkillAFreshness: AClaude Code +9 - 0.561score
harness
dougborg
Audits, bootstraps, updates, and improves a project's agent harness: the skills, agents, hooks, and instructions for Claude Code and Codex.
SkillAFreshness: AClaude Code +9 - 0.558score
implement
dougborg
Build a ticket or small spec in this session, test-first, through to an open PR.
SkillAFreshness: AClaude Code +9
Signals
- Stars
- 0
- Forks
- 1
- Open issues
- 8
- Last commit (this folder)
- Oct 6, 2026 (22 days ago)
- Last release
- Oct 18, 2026 (10 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- Apache-2.0
- Repository
- github.com/dougborg/harness-kit
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from dougborg
- 0.565score
agent-standup
dougborg
Reconciles and coordinates shared work across multiple coding agents and human operators.
SkillAFreshness: AClaude Code +9 - 0.565score
adhd-mode
dougborg
Shapes every response for a reader with ADHD: the next action first, numbered steps, progress restated each turn, tangents held back, concrete time estimates…
SkillAFreshness: AClaude Code +9 - 0.565score
code-reviewer
dougborg
Reviews a diff across six dimensions — correctness, design, readability, performance, testing, security — plus the repo's documented standards, code smells…
SkillAFreshness: AClaude Code +9 - 0.565score
codebase-design
dougborg
Shared vocabulary and principles for designing deep modules: a lot of behaviour behind a small interface, at a clean seam, testable through that interface.
SkillAFreshness: AClaude Code +9 - 0.565score
commit
dougborg
Creates conventional commits with quality gates — stages intentionally, checks for uv.lock drift, runs the project's verification command, and writes the…
SkillAFreshness: AClaude Code +9 - 0.565score
diagnosing-bugs
dougborg
A disciplined diagnosis loop for hard bugs and performance regressions: build a tight feedback loop that goes red on this bug, minimise, hypothesise…
SkillAFreshness: AClaude Code +9
Cite this page
documentation-writer skill. Agent Mart. https://agentmart.dev/skills/dougborg-documentation-writer (last updated 2026-10-24)