Content
46%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 lays out a sensible four-step workflow with real tool names and a repo list, but it is not self-sufficient: the referenced template and classification-rules files are missing from the bundle, the search call lacks the parameters needed to execute it, and a duplicated resources section wastes tokens. The batch search-and-aggregate process also has no validation or empty-result handling.
Suggestions
Actually ship references/classification-rules.md and references/weekly-template.md (or inline the template), since the final output step currently points at files that do not exist in the bundle.
Add the concrete search invocation: the query string for toolbase_github_search_issues including author login, type:pr, and the 7-day date window, plus how results map into the classification step.
Remove the duplicated "可用资源" section and its trailing meta-note, and add a validation checkpoint in Step 2 (e.g., what to do when a repo returns zero PRs or the API call fails).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The workflow steps are lean with no concept over-explanation, but the entire "📦 可用资源" section is duplicated verbatim and closes with a stray meta-note ("根据 SKILL.md 中的 IF-THEN 规则判断是否需要加载") that is noise rather than instruction. Mostly efficient but includes unnecessary content that should be tightened — anchor 3; not 4 because the duplication is a real defect, not minor trimming. | 3 / 5 |
Actionability | Concrete tool names are given ("调用 toolbase_github_get_my_user，记录 login 字段", "调用 toolbase_github_search_issues") plus a repo list and classification keywords, but key executable details are missing: no search query/parameter shape (author filter, type:pr, date range), no worked classification examples, and "按 weekly-template.md 输出" names a file without a usable path. Some concrete guidance but incomplete — anchor 3; not 4 because the search and output steps cannot be executed as written. | 3 / 5 |
Workflow Clarity | Steps 1–4 are listed and sequenced (get user → search PRs → classify → render report), but this batch aggregation across five repositories has no validation checkpoints: no handling of empty search results, API/auth failures, or PR-count sanity checks, and no instruction on what to do when classification rules are ambiguous. The rubric's cap for batch operations without validation applies — anchor 3, and it cannot exceed 3; not 2 because the sequence itself is coherent and concrete. | 3 / 5 |
Progressive Disclosure | The body is short with clearly signaled one-level-deep references (classification-rules.md, weekly-template.md), but neither file exists in the bundle — there is no references/ directory at all — so the mandatory output template is unexecutable, and the resource listing is duplicated. Scored against the actual (empty) bundle structure, this fails the core purpose of progressive disclosure — anchor 2; not 3 because the references are not merely unclearly signaled, they are broken. | 2 / 5 |
Total | 11 / 20 Passed |