Content
75%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 tight, actionable recipe body: concrete model IDs, complete prompts, correct phase gating, and practical fallback instructions. Weaknesses are minor — duplicated description text, a redundant in-body trigger keyword section, a missing exact CLI invocation example, and no output validation before chaining images into video.
Suggestions
Include one complete example invocation of 'muapi image edit' (with flags) and a filled-in curl request body so the steps are copy-paste ready.
Add a lightweight validation checkpoint after image generation (confirm the selfie contains the user and the actor) before chaining into the Phase B video.
Drop the verbatim bold restatement of the description and move the 'Trigger Keywords' list into the frontmatter description to trim the body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean: an inputs table, two concrete phases with full ready-to-use prompts, and short agent notes — it explains nothing Claude already knows. Not 5 because it duplicates the frontmatter description verbatim as a bold subtitle and carries a 'Trigger Keywords' section that adds padded, redundant trigger text in the body. | 4 / 5 |
Actionability | Concrete, near-executable guidance throughout: named models ('nano-banana-2-edit', 'kling-o1-image-to-video'), full verbatim prompts, aspect ratios, a curl fallback with polling ('muapi predict wait <request_id>'), and {{placeholder}} substitution rules. Not 5 because no exact copy-paste 'muapi image edit' invocation with flags is shown, and the curl body is a '{...}' placeholder rather than a complete example. | 4 / 5 |
Workflow Clarity | Sequencing is clear with explicit gating: 'If the {{user_image}} is not provided, ask the user' is an input checkpoint, and Phase B is correctly conditioned on 'at least two selfies' and 'Only when requested'. Not 5 because there is no validation step on the generated outputs (e.g., checking the selfie actually contains the user and actor before chaining into the video). | 4 / 5 |
Progressive Disclosure | The skill is a single well-organized file with clear sections (Inputs, Steps split into Phase A/B, Trigger Keywords, Notes) and no bundle files — no nested or buried references exist. Not 5 because it slightly exceeds the ~50-line simple-skill threshold and the Trigger Keywords block duplicates frontmatter responsibilities instead of being trimmed. | 4 / 5 |
Total | 16 / 20 Passed |