Content
88%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.
A strong, experience-derived skill: an ordered validation loop with real checkpoints and a set of gotchas (cache, overlays, ephemeral ids, theme-relative color assertions) an agent would otherwise rediscover slowly. The only meaningful improvement is removing the duplication between the inline ❌ callouts and the Anti-patterns section.
Suggestions
Remove the Anti-patterns entries that repeat inline ❌ callouts verbatim (soft-reload trust, script-wrapping, artifact/dev-server leftovers), keeping that section only for pitfalls not already flagged in place.
Consolidate cleanup guidance into the Cleanup discipline section and reference it from the loop, rather than restating it as anti-patterns.
Consider a one-line "Evidence to report" checklist after step 5 of the loop to make the reporting requirement scannable without reading the full technique sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious gotchas (ephemeral evaluate-injected ids, cache-busted reloads, hover target requirements) and avoids explaining known concepts, but the ❌ callouts are duplicated — the soft-reload warning and cleanup/artifact warnings each appear both inline in their technique sections and again in the Anti-patterns section — which is trimmable without losing anything. | 4 / 5 |
Actionability | Guidance is fully executable: exact MCP tool names (browser_navigate, browser_snapshot, browser_console_messages, browser_press_key Escape), copy-paste browser_evaluate JS for computed-style comparison and theme toggling, and a concrete cache-busting pattern (?v=<timestamp>) — no pseudocode or vague direction. | 5 / 5 |
Workflow Clarity | The validation loop is a clearly sequenced cheap-to-expensive procedure with "Stop as soon as you have a confident answer", per-step assertions ("assert 0 errors ... Do this on every page you validate"), error-recovery feedback loops (Escape on intercepted clicks, re-query from a fresh snapshot after re-render), and evidence-based pass/fail reporting; no destructive or batch operations that would cap the score. | 5 / 5 |
Progressive Disclosure | Sections are well organized with clear headers and nothing that clearly belongs in a separate file is inlined (no bundle files exist, appropriately). It falls short of the 5 anchor — which rewards content appropriately split across well-signaled references and easy navigation — because the skill is ~130 lines and the duplicated inline ❌ callouts vs. the Anti-patterns section slightly muddy navigation. | 4 / 5 |
Total | 18 / 20 Passed |