agentleFS
strands-agentsstrands-agents/box/.agents/skills/docs-reviewer/SKILL.md

docs-reviewer

Skill490 starsChanged 6 days agoApache-2.0View on GitHub

Review a Box documentation draft in docs/user/ or docs/design/ for placement, structure, altitude, readability, terminology, and claim scope, and give a verdict (Ship it, Tighten, or Rethink). Use after docs-writer produces a draft, before a docs pull request, or when asked to "review this draft", "check my docs", or "is this page ready to ship". Read-only.

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 9 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the draft once straight through, as its reader would.

    line 18
  2. Distinctive:

    Run the mechanical check, and use its hits as input to dimension 4:

    line 19
  3. Distinctive:

    Write the review in the output format.

    line 28
  4. Distinctive:

    No section mixes content types.

    line 38
  5. Distinctive:

    No section exists only to fill a template slot.

    line 39
  6. Distinctive:

    Read aloud, each sentence sounds like a person said it.

    line 51
  7. Very specific:

    No test name, no source file, no issue or pull request number, no file.rs:LINE citation, and no link into .agents/drafts/ or …

    line 72
  8. Distinctive:

    Ship it: All dimensions pass, with one minor warning at most.

    line 85
  9. Distinctive:

    If you can't decide between Tighten and Rethink, ask: can the writer fix this in place, or do they need to re-outline?

    line 95
  10. Very specific:

    Check a claim against the code.

    line 127
  11. Distinctive:

    Commit, push, or approve a pull request.

    line 128

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “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

Documentation reviewer

Scope. Whether a draft is in the right place, shaped around its subject, at the right altitude, easy to read, and consistent with the terminology lock, measured against voice-guide.md. You don't check that a claim is correct against the code and its tests. That's the docs-audit skill.

Readability and altitude come first. A page can pass every mechanical rule and still be a page nobody wants to read, and that's a fail.

Procedure

  1. Read the draft once straight through, as its reader would. Note where you had to reread.
  2. Run the mechanical check, and use its hits as input to dimension 4:

    .agents/skills/docs-reviewer/check-prose.sh <draft>
  3. Classify the content type from the directory and the structure.
  4. Score each dimension below as pass, warning, or fail.
  5. Give one verdict.
  6. Write the review in the output format.

Dimensions

1. Placement and shape

  • The page is in the right guide. A user page that argues a design, or a design page that's a list of options, is in the wrong guide.
  • The structure follows the subject. A page about several parts has one section per part, and no part is spread across sections. A page about one task follows the reader's order.
  • No section mixes content types. Conceptual background in a how-to guide is a finding.
  • No section exists only to fill a template slot.
  • The page doesn't do the job of decisions.md, terminology.md, or tenets.md.

2. Altitude

  • Each detail belongs to this page. Detail that belongs to a sibling topic (a wire protocol on a page about binaries, Seatbelt rules on a page about policy) is a finding, even when that page doesn't exist yet.
  • Length fits the subject. A draft that's much longer than the subject needs is a finding.

3. Readability

  • Read aloud, each sentence sounds like a person said it. Quote each one that doesn't, and give a plainer version.
  • Each section opens with what the reader needs first. A design section says what the thing is before what it guarantees.
  • A list isn't walked through twice.
  • Each term is introduced before its shorthand is used.

4. Constraints and terminology

  • The words not to use, em-dashes, emoji, padded triads, and negative parallelism. Apply the override table for the content type.
  • User guide: describe what is, and no "why" unless it changes the action.
  • Each word from the "do not use" column of docs/design/terminology.md is a finding. A term the lock doesn't have, or a lock term that the code calls something else, is a question for the user.

5. Claim scope

  • Each reason links its decision anchor the first time it comes up, and the decision is summarized, not copied. A reason with no link is a finding.
  • A claim stated more broadly than the mechanism it describes is a finding.
  • No test name, no source file, no issue or pull request number, no file.rs:LINE citation, and no link into .agents/drafts/ or .agents/explorations/.

6. Examples, diagrams, and standing alone

  • Each example is complete and realistic, one concept per block, with the output below the command.
  • Prose and code agree.
  • Each diagram shows the main path only, is Mermaid with every label quoted, and has no ASCII art.
  • The first paragraph says what the page covers. Prerequisites are stated, terms are defined or linked on first use, and no cross-reference carries the load.

Verdicts

Ship it. All dimensions pass, with one minor warning at most.

Tighten. Two or more warnings, or one fail that an edit in place fixes. Typical causes: a few sentences that don't read aloud, detail that belongs on another page, terminology slips, or a draft much longer than it needs to be. Give a fix for each finding, at its line.

Rethink. Two or more fails, or one structural fail: the wrong guide, a shape that spreads each part across sections, a page mostly at the wrong altitude, or prose that reads like a spec throughout. Give the diagnosis and the right approach.

If you can't decide between Tighten and Rethink, ask: can the writer fix this in place, or do they need to re-outline? In place is Tighten. Re-outline is Rethink.

Output format

## Review: <page title>

**Guide and type:** <user guide or design guide>, <type>
**Verdict:** Ship it | Tighten | Rethink

| Dimension | Score | Key finding |
|---|---|---|
| Placement and shape | ... | ... |
| Altitude | ... | ... |
| Readability | ... | ... |
| Constraints and terminology | ... | ... |
| Claim scope | ... | ... |
| Examples, diagrams, standing alone | ... | ... |

### Findings

1. <file>:<line>: <the finding>. Fix: <the fix>.

### What works

<Two or three things the draft does right.>

What this skill doesn't do

  • Edit the draft. The writer fixes it.
  • Check a claim against the code. That's docs-audit.
  • Commit, push, or approve a pull request.

Download now (ZIP)

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. strands-agents/sdk-python★ 9ksame name
    2. strands-agents/harness-sdk★ 8.9ksame name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository11 other files it gives its agents.

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.