Content
92%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 exemplary lean skill file: executable commands covering the common cases, an explicit safety workflow (authorize → preview → confirm → verify), and hard platform constraints stated without padding. The only notable gap is that the deferred 'methods' reference for full comment-operation parameters is not shown as a runnable command.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~13-line body is dense and operational with zero padding: "登录由 youtube-hunter login/export 完成。先确定真实目标、文案与用户授权;默认预览,已有动作授权时加 --confirm" — every sentence adds non-obvious, tool-specific constraints (channel requirement, own-comment-only deletes, replies hang off the root comment, no DMs/gifting). No concept Claude already knows is explained, matching 'every token earns its place'. | 5 / 5 |
Actionability | The bash block gives copy-paste-ready commands with clearly marked placeholders for all common cases (like, follow, comment, reply, delete): "youtube-interact comment '<视频链接>' --text '评论正文' --account default --confirm". Not a 5 because full parameters for comment add/delete/like/unlike are deferred to an unspecified "methods" view without showing that command, and login is delegated to youtube-hunter without an example — minor gaps. | 4 / 5 |
Workflow Clarity | For inherently risky outward-facing actions, the workflow has explicit checkpoints and a feedback loop: sequence is login (youtube-hunter login/export) → "先确定真实目标、文案与用户授权" → default preview with "--confirm" only after authorization → "结果未知先查评论或平台页面,不直接重发" (verify before retry). Replies also require "引用 hunter 读取的真实评论 ID", an explicit pre-action validation. This matches the anchor 'clear sequence with explicit validation steps; feedback loops for error recovery'. | 5 / 5 |
Progressive Disclosure | The skill is under 50 lines with no need for external references (references/, scripts/, assets/ are absent), and the body is well organized: auth/principles paragraph → executable examples → capability map ("like/unlike 点赞/取消;collect/uncollect…") → platform constraints → recording via bd-record. Per the rubric's simple-skill exception, this earns a 5 with just well-organized sections. | 5 / 5 |
Total | 19 / 20 Passed |