Content
90%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.
The body is an exemplar of lean, dense operational writing: fully executable commands, explicit validation gates, and a clear segmentation/merge sub-workflow. The only gaps are unspecified recovery paths on failure and no pointer from the body to the bundled scripts/avatar.py.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every line carries operational constraints (pixel/format limits, 0.75s tail silence, ≤0.12s timing tolerance, 20s/15MB caps) with zero padding or explanation of concepts Claude already knows; the legacy '.liveportrait.json' compatibility note is a legitimate old-pattern remark, not version fluff. | 5 / 5 |
Actionability | Copy-paste-ready commands cover every common case — generate/resume/review/deck-compose and the segment + assemble flow — with concrete flags, defaults, and ranges (--resolution 480P|720P, --timeout 1..3600, --tail 0..2, --start/--duration) plus a literal manifest JSON example. | 5 / 5 |
Workflow Clarity | The pipeline is sequenced (generate → sample 5–10s clip → verify → full render per GATE B → review → assemble → review → compose) with explicit validation gates (check_pass, timing_pass, lip_sync_review_pass, hash binding, per-segment PCM checks). It falls short of 5 because error recovery is only negative guidance — '已确认失败也不自动重提' and '不定帧、不循环、不变速' state what not to do without saying what to do next. | 4 / 5 |
Progressive Disclosure | The body is well organized (intro, canonical commands, constraint bullets, dedicated segmentation section, labeled external Aliyun doc links), but the bundled scripts/avatar.py that implements the `avatar` CLI is never referenced from SKILL.md, leaving a minor navigation gap between the overview and the bundle. | 4 / 5 |
Total | 18 / 20 Passed |