Content
73%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-engineered operational skill: the 12-step workflow is unambiguous, packed with executable commands, and guarded by multiple confirmation gates and error-recovery loops. Its weaknesses are repetition of the multi-clip chaining and constraint material across sections, and a monolithic ~460-line body with no bundle files to split reference-heavy detail into.
Suggestions
Consolidate the chained multi-clip pipeline into one place (step 5) and have steps 11–12 reference it instead of restating the sequencing rules and duplicating the ffmpeg stitch command.
Move the Seedance constraints table and regression notes (including the date-stamped 'confirmed 2026-04-10' items) into a single reference file, keeping only the quick-reference table inline, to reduce repetition and isolate time-sensitive facts.
Add one complete example CreateVideoDto JSON payload and one example adapted prompt in step 6/11 so the composition and generation steps are fully copy-paste ready.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with domain-specific operational detail Claude does not already know (Seedance constraints, chaining rules, credit math), but it could be tightened: the chained multi-clip pipeline rules appear in step 5, again in step 11 (items 3 and 6), with the ffmpeg stitch command duplicated in step 5 rule 5 and step 12; the constraints table and error-recovery table repeat several entries. Date-stamped facts ("confirmed 2026-04-10", "2026-04-09" log citation) also sit outside any deprecated/old-patterns section. Not 4 because the redundancy is more than minor; not 2 because there is no generic concept explanation or padding — nearly everything is skill-specific. | 3 / 5 |
Actionability | Highly executable throughout: exact bash invocations (extract-frames.sh call, prerequisites probe, ffmpeg concat with absolute paths), a runnable whisper snippet, a logging JSON template, a concrete decision tree, and per-error recovery actions. Not 5 because step 11's CreateVideoDto is given as a field checklist rather than one complete copy-paste JSON example, and step 6 gives prompt-composition rules without a single example adapted prompt — minor gaps against the fully copy-paste-ready anchor. | 4 / 5 |
Workflow Clarity | Twelve explicitly sequenced steps with validation checkpoints throughout: a prerequisite check, a user confirmation gate at step 4 ("Wait for user confirmation"), a mandatory dialogue gate at step 7 with re-present-until-approved loop, credit confirmation at step 9 ("Do NOT generate until the user confirms"), a sanity probe for the known 500 regression, and an error-recovery table with per-failure remediation. This matches the anchor of clear sequence, explicit validation, feedback loops, and checklists for a complex process. | 5 / 5 |
Progressive Disclosure | References are one level deep and clearly signaled with purpose ("seedance-2.md — Seedance 2.0 platform rules (read before composing any prompt)", five prompt-library templates, reference.md, main SKILL.md), and the inline constraints table appropriately summarizes while pointing to reference.md for full details. Not 5 because no bundle files exist and the ~460-line body inlines substantial material (the full chaining pipeline, cost tables, prompt-library selection guidance) that could live in a reference file; not 3 because what is inline is the core workflow and navigation to external detail is easy via the Related files section. | 4 / 5 |
Total | 16 / 20 Passed |