Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is highly actionable with a well-sequenced, validated workflow and excellent one-level-deep progressive disclosure of a real bundle. Its main weakness is conciseness, as schemas and the workflow are restated across multiple sections.
Suggestions
Collapse the duplicated per-scene JSON schemas and the three near-identical '使用示例' workflow restatements into a single canonical example plus a compact parameter table.
Remove the '注意事项' restatement of principles already covered in '操作步骤', or replace it with a short cross-reference back to the relevant step.
Trim the full worked output example to one representative scene with a pointer to assets/examples/sample-prompts.md for the complete set.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient and domain-specific, but repeats the per-scene JSON schema, the full worked example, and restates the workflow/principles across "操作步骤", "注意事项", and three "使用示例" blocks that could be tightened. | 2 / 3 |
Actionability | Provides copy-paste-ready, fully specified guidance: an explicit prompt formula, concrete JSON output schemas for every step, a verification checklist, and a worked example with real Chinese/English prompt text. | 3 / 3 |
Workflow Clarity | The five-step sequence is clearly numbered with an explicit quality-check validation step, a pass/fail checklist, and a defined validate→fix→retry feedback loop capped at 2 iterations. | 3 / 3 |
Progressive Disclosure | A clear resource index points one-level-deep to six verified reference files and one asset, each annotated with a '何时读取' (when to read) trigger, keeping heavy detail (scene templates, role definitions) appropriately offloaded. | 3 / 3 |
Total | 11 / 12 Passed |