Content
73%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 an exceptionally rigorous, actionable QA workflow - explicit steps, copy-paste commands, complete reviewer prompts, and a hard completion gate with bounded retry loops. Its weaknesses are token efficiency (a long harness-compatibility preamble and repeated rules that add no visual-QA value) and inlining the full prompt templates and platform-routing boilerplate instead of moving them to reference files.
Suggestions
Move the "Codex Harness Tool Compatibility" preamble (including its thrice-repeated "this section wins" rule and the dense V1/V2 routing paragraph) into a separate reference file or a routing table one-tenth the length; it is generic harness boilerplate, not visual-QA guidance, and consumes ~25 lines of every load.
Extract the two full oracle prompt templates (Pass A and Pass B, ~110 lines) into references such as references/pass-a-prompt.md and references/pass-b-prompt.md, keeping a short charter summary plus the file pointer in SKILL.md.
State recurring rules once instead of three times: the fresh-evidence requirement is repeated across "Evidence must be fresh", capture hygiene, and the completion gate, and could be consolidated into a single up-front rule the later sections reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The visual-QA core (steps, checklist, gates, commands) is dense and operational with no concept over-explanation, but the ~25-line "Codex Harness Tool Compatibility" preamble is harness-routing boilerplate unrelated to the skill's purpose, "If a code block below conflicts with this section, this section wins" is repeated three times, and the V1/V2 paragraph restates routing rules heavily. This is more than the minor trimming of anchor 4, so it fits "mostly efficient but could be tightened". | 3 / 5 |
Actionability | Provides copy-paste-ready commands ("node \"$SKILL_DIR/scripts/visual-qa.mjs\" image-diff <reference.png> <actual.png>", tui-check, pdftoppm, the scroll-owner probe, the web-terminal QA script) and two complete fill-in reviewer prompt templates. Not 5 because the primary web-capture path and machine pre-check depend on external unbundled skills (omowright via the `browser` skill, the `frontend` detect SDK), so the flow is not fully executable from this bundle alone. | 4 / 5 |
Workflow Clarity | Steps 1-6 are clearly sequenced with hard validation checkpoints: fresh-evidence staleness rules, capture hygiene verification before dispatching reviewers, a machine pre-check, and a bounded completion gate with explicit feedback loops (defect -> fix -> re-capture -> FRESH reviewer) and a NEEDS HUMAN REVIEW terminal state on budget exhaustion. This matches the anchor-5 pattern of explicit validation, error-recovery loops, and a checklist for a complex process. | 5 / 5 |
Progressive Disclosure | Bundle structure is good: references/browser-setup.md and scripts/visual-qa.mjs are real, one level deep, and clearly signaled from the body ("$SKILL_DIR/references/browser-setup.md has the matrix loop..."), with capture detail and tooling appropriately split out. Not 5 because ~110 lines of oracle prompt templates and the Codex harness preamble are inlined in SKILL.md where they clearly belong in reference files. | 4 / 5 |
Total | 16 / 20 Passed |