agentleFS
T0mSIlverT0mSIlver/skills/docs-review/SKILL.md

docs-review

Skill1 starsChanged 33 days agoMITView on GitHub

Write, edit or review a project's user docs: README, docs/ pages, CLI, server and plugin READMEs. Checks every claim against the code, gives each fact one page that the others link to, says one behaviour in one sentence wherever it appears, and trims. Use whenever a change touches docs or makes a doc claim stale, and for a docs pass after a release. Triggers: editing README.md or docs/, a renamed flag, label, default or limit, 'refresh the docs', 'review the docs'.

25 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use code-review.

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.
  • GoogleSkill

    code-review

    Adversarial systems, concurrency, memory-model, and architecture code review protocol. Use this skill whenever asked to review code, PRs, diffs, or branches in Filament.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 24 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read the project's own docs guide first (docs/AGENTS.md, a docs section in AGENTS.md or CONTRIBUTING.md).

    line 13
  2. Distinctive:

    Check each claim against the code,: never against another doc, a spec or memory.

    line 23
  3. Distinctive:

    Say each behaviour in one sentence: When two products, platforms or modes behave the same, write the sentence once and put the differences …

    line 32
  4. Very specific:

    Search the repo, the apps and the landing page for each renamed anchor.

    line 41
  5. Very specific:

    Keep an old one that outside pages use with <a id="old-anchor"></a> above the heading.

    line 42
  6. Distinctive:

    Every number, label, flag and path you touched is in the code as written.

    line 92
  7. Very specific:

    Every anchor you renamed is found nowhere else, or kept with an id.

    line 95

“Common” is measured: kept by at least 6 in 10 of the 24 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

Docs review

Docs fail in three ways: a claim the code no longer backs, the same fact told twice in two wordings, and text the reader can't act on. This skill catches all three. For how sentences sound, run unnarrate and then unslop on every paragraph you write; this skill doesn't repeat them.

Read the project's own docs guide first (docs/AGENTS.md, a docs section in AGENTS.md or CONTRIBUTING.md). Its page map, terms and voice win over this skill.

Happy path

  1. List what changed. For a pass after a release, read git log --oneline <previous tag>..origin/main and the decision records (spec/, ADRs) changed since. For a code change, search the docs for each name, flag, label, default and number the change touches.
  2. Check each claim against the code, never against another doc, a spec or memory. Label: search the UI source for the exact string. Number (timeout, size, limit, default): find the constant. Flag: the CLI's help text. Version: the release list. Keep the file and line for each claim you change; the PR body cites them.
  3. Give each fact one home. Each page answers one reader question, and the fact lives on the page whose question it answers. Elsewhere, a few words and a link. When you find a fact on two pages, keep it on its home page and cut the other copy to a link.
  4. Say each behaviour in one sentence. When two products, platforms or modes behave the same, write the sentence once and put the differences beside it. Where the same behaviour must appear on several pages, copy the sentence word for word. A reworded sentence reads as a different behaviour.
  5. Trim. Cut what the reader can't act on: internals, history, why a decision was made (that belongs in the decision record). Paragraphs run about 3 lines.
  6. Keep inbound links working. Renaming a heading changes its anchor. Search the repo, the apps and the landing page for each renamed anchor. Keep an old one that outside pages use with <a id="old-anchor"></a> above the heading.
  7. Run unnarrate, then unslop, on the diff.
  8. Write a short PR body: each stale claim with the file that proves the new one, what moved where, and the word count before and after.

Gotchas

  • Two wordings drift apart. A fact restated on a second page doesn't get the next fix. Rule: link to its home instead. Repeat only the one sentence that names a shared behaviour, verbatim.
  • Terms defined page by page. Readers meet a term on whichever page they land on. Rule: define each term once, on the first page, and link there (one-term keeps one term per thing). Keep a list order, such as the order of supported platforms, the same everywhere.
  • A ✓ with the catch in a footnote. Readers trust the cell. Rule: put the condition in the cell ("CLI only", "Opt-in", "With add-on").
  • Claims wider than the code. "Works in X" when only X's CLI was checked. Rule: name the surfaces checked, and say "untested" for the rest.
  • Claims about other tools age silently. Rule: version or date each one ("CLI 0.160 or later", "menu names from October 2026"), and link that tool's docs instead of copying them.
  • Versions in examples. A pinned v1.2.3 in a clone or upgrade command goes stale every release. Rule: link the latest release, show the tag as an example, and update it on every pass.
  • "Since 1.4" notes. Once every supported release has the feature, the note is noise. Rule: drop it.
  • UI labels from memory. Paraphrased labels send readers hunting. Rule: copy the exact string from the source and give the path, such as Settings → Notifications → Sound. Defaults and choices come from the code too: a setting the docs call opt-in may be on by default.
  • A feature that changed shape. A switch that became a choice, or a "comes back at the end" that now has a mid-way path. Rule: for each release change, reread every sentence that describes the old behaviour, not only the one the PR touched.
  • Maintainer content in the user path. Contract tables, CI internals and schemas stop readers where they sit. Rule: move them below the user content, or to their own page.
  • No proof at the end of setup. Rule: the getting-started steps end with one that shows it works (a test message) and say what to run when it doesn't. List every file and service setup writes on the machine.
  • Reviewer rewrites of picked words. Rule: a label, name or term the owner picked stays. A reviewer's rename is a proposal for the owner, not an edit.
  • Docs read by agents. Each always-loaded line costs tokens on every turn. Rule: keep the steps in the main file, and move reference that only some readers need behind a link.

Checks before the PR

  • Every number, label, flag and path you touched is in the code as written.
  • Each fact you touched lives on one page; others link to it.
  • Each shared behaviour reads the same sentence everywhere.
  • Every anchor you renamed is found nowhere else, or kept with an id.
  • A fresh reader, such as a model with no repo access given only the changed pages, can follow the steps to the end (reference/sources.md, "Reader test").

Not possible

  • Checking another tool's UI without running it. Say "checked on <date> with <version>" only for what you ran; for the rest, link that tool's docs.

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.