Content
88%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.
An exceptionally actionable and well-validated operational protocol: exact tool contracts, sequenced workflows with verification at every checkpoint, and thorough failure/recovery handling for a browser-automation domain where those matter most. Weaknesses are minor: some rule repetition and overly long sentences hurt conciseness, and everything lives in one long SKILL.md rather than being split across reference files. One formatting glitch merges numbered steps 3 and 4 of the browserTask section into a single line, slightly obscuring the list structure.
Suggestions
Deduplicate the draft_not_found retry rules, which are stated in full both in “从视频工程自动导出并发布” and in “API 优先与路由” — state them once and cross-reference.
Move the seven per-action browserAction procedures (search-videos, list-videos, video-data, list-comments, publish-draft, comment-video, reply-comment) into a references/ file per action, keeping SKILL.md as the routing/overview layer.
Fix the formatting glitch where numbered steps 3 and 4 of the browserTask execution section are merged onto one line, and break up the longest run-on paragraphs into bulleted rules for scannability.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every sentence carries plugin-specific operational rules Claude could not know (“无需安装额外依赖或新建执行器” is the only setup note) with no explanation of known concepts. Not 5: several sections repeat the same rules (draft_not_found retry logic appears in both “从视频工程自动导出并发布” and “API 优先与路由”), and very long run-on sentences (e.g., the browserTask step 4 paragraph) could be tightened. | 4 / 5 |
Actionability | Fully concrete guidance: exact tool signatures like “save-draft(accountId,title,text,assetId?,id?,runKey?)” and “finish-browser-job(jobId,executionToken,actualProfileId,actualAccount,outcome,evidence,publicationStatus?,resultUrl?,items?)”, a concrete call example with literal args (“ipollowork_extension_call(extensionId="media", action="video_render_start", args={sourcePath:...})”), and precise sequences (“import-media → save-draft(新建不传 id)→ publish-draft → browserTask 时领取并执行 → finish-browser-job → get-job”). Not 4: commands and parameters are copy-paste ready across all common cases. | 5 / 5 |
Workflow Clarity | Multi-step processes are clearly sequenced with explicit validation checkpoints and error-recovery loops: “调用 get-job 验证已保存,再报告实际结果”, mandatory identity verification before writes (“网页写操作每次都重新核对当前登录身份”), the draft_not_found one-retry rule, the closed-tab recovery loop (“最多2次”), and strict uncertain/failed/pending outcome handling. Not 4: checkpoints and feedback loops are explicit at every stage, including batch operations (comment/search limits of 20 items). | 5 / 5 |
Progressive Disclosure | Well-organized sections with clear headers and a coherent single-file structure; no bundle files exist (references/, scripts/, assets/ are absent) so nothing is mis-signaled or nested. Not 5: at ~93 dense lines, per-action detail (e.g., the seven browserAction sub-procedures) is all inlined where a references/ split per action would keep SKILL.md closer to an overview; not 3 because structure is good and navigation between sections is easy. | 4 / 5 |
Total | 18 / 20 Passed |