Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is highly actionable and well-structured, with concrete muapi commands, a clear two-phase workflow gated by a user pick, and explicit failover recovery. The main weakness is mild verbosity — a repeated description lead-in, a redundant agent-notes section, and some rough 'the plan' phrasing — that could be tightened without losing clarity.
Suggestions
Remove the redundant bold restatement of the description under the H1, and tighten 'Submit ONE the plan' / 'a SECOND the plan' to 'submit one plan' / 'submit a second plan'.
Fold the 'Notes for the Executing Agent' curl/poll guidance into the relevant step or a single short 'Fallback' note to avoid repeating orchestration mechanics.
Keep the lean command-table style but ensure inline placeholders like {{duration_sec}} are resolved-before-call is stated once rather than implied across steps.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean and assumes Claude's competence (specific model IDs, parameter lists), but repeats the frontmatter description as both the H1 and bold lead-in, and includes awkward phrasing like 'Submit ONE the plan' / 'a SECOND the plan' plus a somewhat redundant 'Notes for the Executing Agent' section — matching 'mostly efficient but could be tightened'. It is not padded with concepts Claude already knows (so not 1) but is not maximally lean (so not 3). | 2 / 3 |
Actionability | Concrete, executable commands are given throughout — 'muapi image generate', 'enhance_image' (operation=upscale), 'muapi video from-image' (model=kling-v3.0-standard-image-to-video, duration={{duration_sec}}), 'muapi audio create' — with model IDs, parameters, $nX.url wiring, and a curl fallback with polling via 'muapi predict wait <request_id>', matching 'fully executable code/commands; copy-paste ready'. | 3 / 3 |
Workflow Clarity | Two phases are explicitly sequenced with a hard user-pick checkpoint between them ('Wait' after listing asset_ids) and explicit error recovery ('If video gen fails after failover, fall back to a still-frame slideshow'), plus a cost-gated checkpoint ('Don't auto-confirm step 4 — its cost (~80 cr) deserves a user nod'). This matches 'clear sequence with explicit validation steps; feedback loops for error recovery'. The operation is generative (not destructive/batch), so the destructive-cap does not apply. | 3 / 3 |
Progressive Disclosure | The skill is a single self-contained SKILL.md with no bundle files and no nested references, organized into clearly labeled sections (Inputs table, Steps with Phase A/B, Notes, Trigger Keywords, Notes for the Executing Agent). Per the rubric, a short reference-free skill with well-organized sections scores 3; there is no monolithic wall of text or multi-level reference chain. | 3 / 3 |
Total | 11 / 12 Passed |