Content
67%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a well-structured, highly actionable four-phase workflow with concrete breakpoints, tool invocations, and a severity-ranked reporting contract. Its main weakness is redundancy: the exclusion list appears twice (description plus intro plus a dedicated 'When not to use' section), and a small capture-verification checkpoint is missing.
Suggestions
Remove the 'When not to use this' section (and the duplicated negatives in the intro paragraph) — the frontmatter description already carries this exclusion list and is always loaded with the body, saving ~25 lines of tokens.
Add a short copy-paste-ready Playwright fallback script for the no-MCP case, since the specific method calls are already named but not assembled into a runnable command.
Insert a validation checkpoint in Phase 1 (e.g., confirm each screenshot is full-page and non-blank before moving to the checklist) to close the workflow's only sequence gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly checklist-driven and efficient, but the intro paragraph ('not a pixel-diff regression tool, not a conversion/copy audit, not an aesthetic design critique') and the ~25-line 'When not to use this' section restate the frontmatter description's exclusion list, which is always loaded alongside the body. This duplication is more than the 'minor instances' of the 4 anchor, though there is no padded concept explanation that would warrant a 2. | 3 / 5 |
Actionability | Gives exact viewport dimensions ('375x812 (mobile - iPhone-class)', '768x1024', '1440x900'), concrete tool calls ('browser_resize + browser_navigate + browser_take_screenshot with fullPage: true', 'page.setViewportSize()', 'page.screenshot({ fullPage: true })'), a file naming convention, and a model finding ('pricing table overflows viewport at 375px, third column cut off'). Not 5 because no copy-paste-ready fallback script is provided for the no-MCP case. | 4 / 5 |
Workflow Clarity | Phases 0-3 (Scope -> Capture -> Structural checklist -> Findings) are clearly sequenced with defaults, a page-count cap ('More than 8 pages in one run dilutes the review'), a severity table, and an explicit output contract. Not 5 because there is no checkpoint verifying captures succeeded (e.g., full-page rendered, no blank screenshots) before the checklist walkthrough, though the workflow is read-only so the destructive-operation cap does not apply. | 4 / 5 |
Progressive Disclosure | No bundle files exist and the ~130-line body is one well-sectioned file whose 16-item checklist is the core operational content, so inlining is appropriate and navigation is easy. Not 5 because the 'When not to use this' block duplicates content that belongs in the description, keeping the body from being a purely clean overview-plus-operations split. | 4 / 5 |
Total | 15 / 20 Passed |