Content
60%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 well-structured, constraint-rich overview that delegates detail to a real, correctly-scoped bundle of three files with clearly signaled roles. The main deficits are that nothing in the body is directly executable — the Run section is comments-only and the PIL implementations must be sourced from an unshipped external reference — and that the Contract section repeats material already covered, costing tokens without adding guidance.
Suggestions
Make the Run section executable: replace the two comment blocks with at least the actual FFmpeg invocations for center-crop, dice/concat, and the explicit-map mux (including the ffprobe bitrate verification as a checkable command), since those are deterministic and belong in the body.
Deduplicate the Contract section — the FREE/deterministic, one-clip-many-cuts, and real-logo-never-freeze rules each restate "The FREE assembly" and the intro verbatim; keep the Contract as short invariants and drop the repeated rationale.
Reference bundle files with their real paths (scripts/PIPELINE.md, scripts/config.example.json) and either clarify what "README.md — documents the FREE assembly (below)" points to, or point to it by path — and state where the unshipped Soundboks PIL reference implementations can be found, since the skill depends on them.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient — no concept explanations Claude already knows, and every section carries real constraints — but there is noticeable redundancy that could be tightened: "Deterministic, FREE (PIL + FFmpeg), no paid calls" and "never a paid call per segment" each appear twice (intro/Run vs. Contract), and the real-logo end-card requirements ("base64-decoded", "never typeset", "never freeze") are stated in full in both "The FREE assembly" and "Contract". This fits the anchor for mostly efficient with some unnecessary repetition, rather than the noticeably-verbose anchor below it, since nothing is concept-level padding. | 3 / 5 |
Actionability | The body names concrete config keys ("config.brand_voice", "music.bpm", "beat_structure.concat_order", "config.text_cards", "config.end_card"), exact parameters (1080×1920, ~6° skew, 1.08× echo, ±1% scale + ±3px drift, 192 kbps), and one genuinely executable FFmpeg directive ("-map 0:v -map 1:a"). But the "Run" block contains only comments — no runnable command — and the actual PIL implementations live in an external "Soundboks reference" that is "copied + adapted per run, not vendored" and absent from the bundle. This matches the anchor for some concrete guidance with pseudocode instead of executable code and missing key details; it is not the minimal-high-level-hints anchor (2) because the parameter and config-field specificity is high, and not anchor 4 because nothing in the body can actually be executed. | 3 / 5 |
Workflow Clarity | The five phases are clearly sequenced ("Phase 0 assets → Phase 1 PAID Seedance + music (parallel, gated) → Phase 2 FREE PIL cards → Phase 3 FREE dice/intercut/concat/mux → Phase 4 watch/QC") with a documented config-field → step map in PIPELINE.md, and validation checkpoints exist (the ffprobe ≈192 kbps check against the silent ~1 kbps garbage-audio failure mode, plus the Phase 4 watch/QC gate). It falls short of anchor 5 because the error-recovery loop is only implied (no "if ffprobe shows garbage audio, re-mux with explicit maps"-style feedback step) and the phase sequencing in the body is prose, not a numbered, checkpointed list. | 4 / 5 |
Progressive Disclosure | Scored against the actual bundle: all three referenced files exist (scripts/PIPELINE.md, scripts/README.md, scripts/config.example.json), each is announced with a clear one-line scope (config → structure to copy; PIPELINE.md → full field → step map with beat variants; README.md → the FREE assembly), and references are one level deep. The SKILL.md body is a genuine overview with details pushed to the bundle. It misses anchor 5 on navigation precision — files are referenced by bare name without the scripts/ prefix ("PIPELINE.md" rather than scripts/PIPELINE.md), and "README.md — documents the FREE assembly (below)" is ambiguous about location — leaving minor organization gaps that fit anchor 4. | 4 / 5 |
Total | 14 / 20 Passed |