Content
85%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 highly actionable, well-structured orchestration skill: concrete commands, explicit validation gates with feedback loops, and a clean one-level reference layout verified against the actual bundle. The main weakness is redundancy — the real-acceptance mandate and related evidence rules are restated several times across sections, and the dense multi-clause sentences could be consolidated.
Suggestions
State the real-acceptance gate rule once (e.g., in 执行约定) and have other sections reference it briefly instead of restating it with variations in 必须执行的节点, 本地辅助工具, and 对用户的交付.
Break up the longest multi-clause policy sentences (e.g., the 执行约定 bullets packing 4–5 rules each) into shorter single-rule statements so each constraint is scannable and unambiguous.
Consolidate the repeated 'do not treat file existence / install success / static checks as capability or acceptance' warnings into one negative-evidence rule near the acceptance node.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | There is no concept-explaining padding, but the real-acceptance/evidence mandate is restated with variations across 执行约定, 必须执行的节点, 本地辅助工具, and 对用户的交付, and many sentences stack 4–5 policy clauses each, so the body could be meaningfully tightened. It fits the "mostly efficient but could be tightened" anchor better than the "minor instances" anchor above. | 3 / 5 |
Actionability | Copy-paste-ready commands with placeholders are provided verbatim ("python3 <skill-dir>/scripts/project.py init <project-dir> --name <项目名> --idea <用户原话>"), along with concrete flags (--goal local|materials|live, --complexity simple|complex), named sub-skills per node, and concrete deliverable checklists — fully executable guidance covering the common cases. | 5 / 5 |
Workflow Clarity | The continue-from-project section sequences next steps in a table with explicit advancement criteria, and validation is explicit with feedback loops: "只有 `gate_passed: true` 且有可核对的真实操作记录才可关闭依赖交付", "失败已修复复验", and a dedicated diagnosing-bugs node for failures with re-verification in the original scenario. | 5 / 5 |
Progressive Disclosure | The body is a clear overview with well-signaled one-level-deep references (all 10 cited references/*.md, required-skills.json, scripts, and the starter manifest.json actually exist in the bundle), each linked in context at the point of need; reference files only cross-link laterally within references/, not in nested content chains. | 5 / 5 |
Total | 18 / 20 Passed |