Content
78%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 well-structured and highly actionable, with concrete commands, model IDs, and a fallback path. Its main weakness is the absence of an explicit validation/feedback loop for generation steps, leaving workflow clarity just short of excellent.
Suggestions
Add an explicit polling/wait step (e.g., `muapi predict wait <request_id>`) into each generation phase rather than only mentioning it in the Notes, plus a retry-on-failure loop.
Show the full `muapi image edit` invocation with flags rather than just the subcommand and model, to make Phase A fully copy-paste ready.
Drop the bold intro line that restates the description verbatim to remove redundancy and save tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is efficient and assumes Claude's competence, with only minor trim opportunities such as the duplicated intro line restating the description and the somewhat long example dialogue script. | 4 / 5 |
Actionability | Provides concrete guidance — specific `muapi image edit` calls, model IDs (gpt-image-2-text-to-image, sd-2-omni-reference, veo3.1-image-to-video), copy-paste prompts, and a curl fallback — with minor gaps in showing the full CLI invocation form. | 4 / 5 |
Workflow Clarity | A clear two-phase sequence with an explicit image-approval checkpoint, but it lacks an integrated poll/wait step and an error-recovery feedback loop for generation failures. | 4 / 5 |
Progressive Disclosure | A self-contained, single-purpose skill with well-organized sections (Inputs, Phased Steps, Trigger Keywords, Notes) and no nested references; the simple-skill exception applies since no external files are needed. | 5 / 5 |
Total | 17 / 20 Passed |