Content
81%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, highly operational skill body: executable commands, hard rules, a validated workflow with mandatory QA and retry loops, and a well-signaled read order. The main deductions are minor: duplicated Read order/See also content, script-internal upload detail that could be trimmed, a stale "One of the 5 above" aspect-ratio reference, and cross-bundle references that don't resolve inside the skill directory itself.
Suggestions
Fix the stale count in the Inputs table: "Aspect ratio | One of the 5 above" should read "One of the 3 above" to match the supported set (1:1, 16:9, 9:16).
Deduplicate the "Read order" and "See also" sections — five links appear in both; keep the numbered read order and reduce "See also" to the non-workflow links only.
Trim the Configuration section's presigned-upload mechanics (single-use filePath, per-variant re-upload) to a one-line note that the script handles uploads internally, since the user only ever passes local paths.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and almost entirely operational (endpoints, auth, hard rules, modes, error table), but has minor trimmable fat: the "Read order" section duplicates five of the six "See also" links, and Configuration explains presigned-upload internals ("Each filePath is single-use — the script re-uploads per variant") that the script already handles since the user just passes local paths. | 4 / 5 |
Actionability | Phase 5 gives copy-paste-ready bash invocations for both modes, a curl health-check, a common-errors table with fixes, and concrete input tables. Minor gaps: the Aspect ratio row says "One of the 5 above" while only 3 ratios are supported (stale count), and command paths mix ~/.claude/skills/... absolute form with relative scripts/... form. | 4 / 5 |
Workflow Clarity | Seven clearly sequenced phases with explicit mandatory checkpoints: Phase 4 blocks on credit-cost confirmation, Phase 6 mandates per-variant visual QA with a defect list and a capped retry loop ("regenerate with a revised prompt explicitly correcting the issue... Cap at 2 retries per variant"), and Phase 7 confirms handoff with the user. This matches the top anchor including feedback loops for a batch/paid operation. | 5 / 5 |
Progressive Disclosure | Good structure: a numbered "Read order" that signals exactly when to open each reference, and one-level-deep links to the shared prompt library, guide, safety suffixes, and the bundled scripts/generate_image.py (which exists and matches the documented contract). Not a 5 because the shared/* and sibling-skill references point outside the skill bundle (../../shared/..., ../arcads-external-api/...) and are not present in the actual bundle, slightly weakening navigation. | 4 / 5 |
Total | 17 / 20 Passed |