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 dense, highly actionable skill body with real executable commands, explicit validation checkpoints, and well-signaled one-level references. Its main weaknesses are repetition, inline policy/pricing detail that belongs in a reference file, and one bundled reference (examples.md) that is never linked from the body.
Suggestions
Deduplicate repeated guidance (mode-selection semantics stated twice; 'never create replacement tasks' stated three times) into single statements at first mention.
Move the pricing, promotion, refund, and file-retention policy in 'Time packs and file retention' into a reference file (e.g. capabilities.md or a new billing.md) and keep only the operationally relevant summary inline.
Link references/examples.md from the body where prompt examples would help (e.g. from the prompt-writing or workflow sections) so the bundled file is discoverable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most content is product-specific operational fact Claude cannot know (quota arithmetic, request-ID semantics, splitting rules), but there is noticeable repetition — 'Mode selection labels the task and does not add directions' appears twice (lines 101 and 110) and 'never create replacement tasks' is stated three times (lines 112, 114, 127) — and dense policy detail (USD 9.90/29.90/79.90/199.90 pricing, refund and promotion semantics in 'Time packs and file retention') sits inline where a reference file would do. Not a 2 because there is no padding explaining concepts Claude already knows; not a 4 because the repetition and inline policy sections could meaningfully be tightened. | 3 / 5 |
Actionability | Fully executable, copy-paste-ready commands cover the common cases: 'node scripts/foleyix.mjs help / whoami / login --no-browser / quota --json', a complete 'generate --mode narration --prompt "..." --out ./welcome.wav --json' example, plus status, download, voices, and jobs invocations, alongside concrete tables mapping modes to required input. Score 5 per anchor: specific examples cover the common cases. | 5 / 5 |
Workflow Clarity | The body opens with a routing section ('Choose the requested workflow') and each workflow is sequenced with explicit checkpoints: check 'quota' before generating, verify 'the result's status and absolute file path before calling it complete', and 'Stop on an unknown result, failure, insufficient balance/storage or a wait timeout'. Error-recovery feedback loops exist (retry with the same persistent request ID, resume with --request-id reusing completed parts), so the batch long-script workflow is not missing validation and the destructive/batch cap does not apply. Not a 5 because the full generation sequence is scattered across several sections rather than given as one ordered checklist, and listening verification is only conditionally encouraged ('when available'). | 4 / 5 |
Progressive Disclosure | References are one level deep, clearly signaled with markdown links, and each is scoped to a purpose (prompt-writing.md for prompts, cli.md for CLI details, direction.md for directing, capabilities.md for limits, speech.md / sound-design.md for content branches, audio-capabilities.json for the catalog) — all cited paths exist in references/. Not a 5 because references/examples.md is bundled but never linked from the body (a discoverability gap), and some content that belongs in a reference (pricing, refund and retention policy) is inlined in the main file. | 4 / 5 |
Total | 16 / 20 Passed |