Content
68%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 efficiently structured, largely actionable recipe with clear phase sequencing and sensible user-confirmation checkpoints. Its main gaps are the absence of result-validation/retry steps for the batch scene and video generation, and prompt templates that still contain bracketed placeholders rather than fully executable examples.
Suggestions
Add validation checkpoints between phases, e.g. 'After each generation, run `muapi predict result` and retry or adjust the prompt if the output failed or lost character consistency' — this would lift workflow_clarity above the batch-operation cap of 3.
Replace the three near-identical per-scene prompt blocks with one parameterized template plus a short scene-beat table (beginning/middle/end) to remove repeated boilerplate and improve conciseness.
Include one complete executable example of a `muapi image generate` / `muapi image edit` command with real argument syntax so guidance is copy-paste ready.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body uses an efficient inputs table, templated prompts, and short phase descriptions with no explanation of concepts Claude already knows. Minor trimmable redundancy remains: the bold restatement of the description under the H1, the repeated 'Cinematic lighting, Pixar 3D animation style' suffix in every prompt, and the vague 'generated in parallel or sequentially depending on the story flow' note keep it below 5. | 4 / 5 |
Actionability | Concrete guidance is present throughout: named models ('nano-banana-pro', 'flux-kontext-pro-i2i', 'kling-v3.0-pro-image-to-video'), aspect ratios, full prompt templates, and a curl fallback with polling. It is not fully copy-paste ready — prompts contain bracketed '[Describe the beginning of {{story_premise}}]' placeholders and no complete 'muapi' command example with argument syntax is shown. | 4 / 5 |
Workflow Clarity | Phases A/B/C are clearly sequenced with user-confirmation checkpoints, but this is a batch operation (3 scenes plus 3 videos) with no validation or retry steps — nothing instructs checking that generation succeeded (e.g. via 'muapi predict result') before animating. The batch-without-validation guideline caps this at 3 even though the sequence itself is well laid out. | 3 / 5 |
Progressive Disclosure | A single-file skill (~85 lines) with well-organized sections (Inputs, phased Steps, Trigger Keywords, agent notes) and no nested references, so structure is good. It is not 5 because the repetitive per-scene prompt templates are inlined where a small references file (or a single parameterized template) would tighten the main file. | 4 / 5 |
Total | 15 / 20 Passed |