Content
81%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 dense, well-structured operational skill: clear routing tables, exact session/profile conventions, and a numbered search workflow with explicit validation, bounded budgets, and error-recovery rules that exceed typical checkpoint coverage. The main gap is that every detailed reference (17+ sites/*.md files and site-catalog.json) points outside the file, and none of those files are present in the reviewed bundle.
Suggestions
Ship the sites/*.md rule files and site-catalog.json alongside SKILL.md in the bundle — the body's core value (per-site URL construction and search rules) lives entirely in files that are absent here, leaving the URL-construction step unactionable as written.
Add one complete end-to-end example for the most common case (e.g., a concrete Bing search URL with an encoded Chinese keyword, the open command, and the expected snapshot/ref extraction) so the primary flow is copy-paste ready without opening a site file.
Consolidate the hunter-skill routing rule into its dedicated table and drop the repeated caveats from the session and site-selection tables (cross-reference the section instead) to reclaim tokens without losing clarity.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense operational rules and tables with no concept-explanation padding — nothing teaches Claude what a browser or search engine is. Not 5 because the hunter-routing rule is restated about four times (dedicated section, session table caveats, site-table rows, closing guidance) and the --persistent rule is repeated, so it could be tightened; not 3 since the restatement is contextual reinforcement at decision points rather than unnecessary explanation. | 4 / 5 |
Actionability | Concrete, executable guidance throughout: exact profile names (default/douyin/xhs-browse/twitter/weibo/tiktok), exact flags, a real command ("camoufox-cli --persistent --json open '<按 sites/zhihu.md 生成的完整搜索URL>'"), exact budgets ("每站最多两次搜索导航/翻页"), and specific field rules ("保留小红书 xsec_token 等访问参数"). Not 5: only one command example with a placeholder URL — no complete end-to-end example (e.g., a concrete Bing search URL plus snapshot/extract flow) covering the most common case. | 4 / 5 |
Workflow Clarity | The 7-step 搜索执行 sequence includes explicit validation checkpoints and feedback loops: bounded waits with "不把空骨架判为无结果", "精确日期到原文核实", timeout recovery via "延迟或超时先核对同一页面和会话状态，不自动重启全部浏览器", and closing self-reporting of sites, counts, and gaps. Not 4 because checkpoints are explicit and comprehensive rather than 'mostly present'; the destructive/batch cap does not apply to a read-only search skill, and the pagination batch is explicitly bounded. | 5 / 5 |
Progressive Disclosure | Good structure: the site-selection table clearly maps each search goal to one-level-deep sites/*.md rule files with descriptive purposes, and site-catalog.json centralizes the six new site categories. Not 5 because the referenced files (sites/*.md, site-catalog.json) are absent from the bundle being reviewed, so the reference structure cannot be verified as real, and browser-guide is referenced without a resolvable path. | 4 / 5 |
Total | 17 / 20 Passed |