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 a lean, well-structured, instruction-style recipe with concrete models, prompts, and fallback commands, and an explicit error path for unreachable URLs. Its main gaps are the missing concrete muapi CLI invocation syntax and the absence of a post-generation verification checkpoint, plus minor token duplication of the description.
Suggestions
Show one complete `muapi image generate` invocation with actual flags (e.g. `muapi image generate --model gpt4o-text-to-image --aspect-ratio 9:16 --prompt "..."`) so the calls are copy-paste ready.
Add an explicit post-generation checkpoint: verify both request results succeeded (via `muapi predict wait`) and retry or report errors before presenting the mockups.
Remove the bold restatement of the frontmatter description and fold the Trigger Keywords into the description to trim duplicated tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is efficient — an inputs table, concrete prompts with exact model IDs, and operational notes — with no explanations of concepts Claude already knows. It is not level 5 because the bold restatement of the frontmatter description under the title is duplicated framing, and the 'Trigger Keywords' section largely repeats information that belongs in the description, tokens that could be trimmed. | 4 / 5 |
Actionability | Guidance is largely concrete and executable: exact models ('gpt4o-edit', 'gpt4o-text-to-image', 'bytedance-seedream-v4.5'), full copy-paste-ready prompts with documented {{placeholder}} substitution, aspect ratios (9:16, 9:19.5), and a near-complete curl fallback command. It stops short of level 5 because the actual `muapi image generate` CLI invocation syntax (how the prompt, model, and aspect-ratio flags are passed) is never shown, and the curl fallback leaves `<endpoint>` and a literal `{...}` body unspecified. | 4 / 5 |
Workflow Clarity | The sequence is clear — two parallel generation steps with conditional prompt logic (screenshot vs. no screenshot, keep_brand), then post-generation presentation steps — and it includes a real error-recovery path ('If the URL is unreachable, ask the user to provide a screenshot') plus polling via `muapi predict wait`. It is not level 5 because there is no explicit checkpoint verifying that both images generated successfully before presenting them. This is not a destructive or batch operation, so the level-3 validation cap does not apply. | 4 / 5 |
Progressive Disclosure | The body is roughly 50 lines, single-purpose, well-organized into clearly labeled sections (Inputs, Steps, Notes, Trigger Keywords, Notes for the Executing Agent), contains no bundle files and no external references to break or nest — matching the rubric's simple-skill provision that short, well-sectioned skills with no external file needs score 5. It would drop if any reference pointed at a missing file, but none do. | 5 / 5 |
Total | 17 / 20 Passed |