agentleFS
Sign inStart free
FracteraFractera/web4-node-starter-kit/.claude/skills/review-page/SKILL.md

review-page

Skill68 starsChanged yesterdayMITView on GitHub

Build process. Review a finished page of an element against its approved design: screenshots and the element's page guard without a model first, then a verdict in the owner's three grades with named sections — 1 «replace X → Y», 2 «move X and Z», 3 «looks great» — recorded next to the owner's verdict. Load it when the owner asks to assess, review or check a page of an element, a landing after an element agent's work («look at what the agent built», «grade this page»), and before any verdict on whether a page's sections are right. The rule that changes behaviour: a review by a model costs tokens and runs only after the owner's yes with a price counted for this page; the instrument (screenshots, the guard) is free and runs first.

4 other projects use a version of this, 2 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 1

    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:

    No false complaints: on a right page do not invent a replacement for the sake of a grade.

    line 53

“Common” is measured: kept by at least 6 in 10 of the 4 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

review-page

Process: build. A hint, not a law, in HOW you review. Not in the price: a review by a model — only after the owner's yes.

1. The instrument — zero tokens, always first

node .claude/skills/review-page/shots.mjs --element <element folder> --page en [--page ru] --out <folder>
node .claude/skills/review-page/shots.mjs --url http://127.0.0.1:<port> --page en --out <folder>
  • --element serves the element's installed build on a spare port of 127.0.0.1 (range 25500–25679) and stops it at the end; a turned-off element is not turned on. It prints the map «screen → sections» (kind·address) and, verbatim, the element's page guard (scripts/check-page-sequence.mjs).
  • A page goes without the leading slash (en, ru/docs): Git Bash rewrites /en into a Windows path and the instrument refuses it.
  • Screenshots: 1440 and 500 px, by screens of 900 px. The build, not the source: an edit after the build is not on the screenshots.
  • A red guard means the verdict is already grade 1 or 2 by its lines; no model is needed for that.

2. The price — named before the yes

An image 1440×900 ≈ 1 700 tokens, 500×900 ≈ 600. A page of three screens at 1440 and four at 500 ≈ 7–8 thousand per language; the element's DESIGN.md and the approved test page's screenshots ≈ 3 thousand; the rubric below ≈ 2 thousand; the verdict ≈ 1–2 thousand. It is an estimate, not a measurement — say so. A second language adds only its screenshots.

3. The rubric — what to look at on the screenshots

Three sources, in order:

  1. The element's approved design: its DESIGN.md (composition, rhythm, imagery, «do» and «don't») and the screenshots of the approved test page (prototypes/_pages/<id>/shots/ or the core's Design tab) — does this page speak the same language; a section's card (components/sections/<section>.card.json, when the element keeps cards): use and avoid, the page's mode (a landing persuades).
  2. craft-floor of the impeccable skill (.claude/skills/impeccable/reference/craft-floor.md of any element) — «Verify» (contrast, spacing, type, motion, states, language) and «Refuse» (identical cards as the skeleton, an eyebrow over the heading, gradient text…).
  3. critique.md of the same skill — parts only: «Design Specificity Verdict» (made for this product or fits any), «Priority Issues» P0–P3, the mode rule (on a landing heuristics 7 and 10 are n/a). The full critique report is not written. 🛑 Critique wants two subagents; they are forbidden here — the review is always «DEGRADED: single-context», and that is its first line.

4. The verdict — exactly one of three grades, with named sections

GradeWhenWhat is named
1 «does not meet the standard — replace X → Y»the section is not for this place or task; another section of the element, or a new one, is betterX → Y by section names
2 «better, but out of place — move X and Z»the sections are right, their order or neighbours are notwhich sections and where
3 «looks great»no complaint that changes sections— (small things in a line; they do not lower the grade)
  • A complaint about the words inside a right section is not a replacement: it is a line under the verdict.
  • The needed section does not exist yet — grade 1 with X → (new); the element's design pass writes it (a task to its agent: npm run task -- <id> "Build process, design: …").
  • No false complaints: on a right page do not invent a replacement for the sake of a grade.

5. The record

node .claude/skills/review-page/verdict.mjs add --element <id> --page en --grade <1|2|3> --by model:<model id> \
     [--replace "section-a>section-b"] [--move section-x] [--fine section-y] [--note "…"]
node .claude/skills/review-page/verdict.mjs add … --by owner                      # the owner's verdict — next to it
node .claude/skills/review-page/verdict.mjs add … --copy "<what is spoiled>"      # a review of a spoiled copy — not compared
node .claude/skills/review-page/verdict.mjs list [--element <id>]                # how often the model agrees with the owner

The file is data/node/page-verdicts.jsonl of the core (outside git). The owner's verdict is recorded in his words in --note — that measures how far the reviewer agrees with his eye. Fixes after a verdict go as a task to the element's agent, not by your hands.

6. What the skill does not do

It does not review every build by itself (no timers, no autostart) · it does not edit the page · a review on another model in a separate context — only by the owner's word.

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.

Check your setup, free