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 a tight, rule-driven instruction set with concrete rewrite recipes, a validation-backed workflow, and a clean single-level reference. The only weakness is minor internal duplication of rules and pattern lists between the body and the reference file.
Suggestions
De-duplicate the LaTeX/受保护片段 protection list — state it once and cross-reference, rather than listing tokens in both 硬规则 and 受保护片段.
Keep the 整齐二分 pattern enumeration in references/machine-patterns.md and reference it from 风格目标 instead of repeating the full list in the body to save tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient and free of padding about concepts Claude already knows, but several rules are restated (LaTeX protection in both 硬规则 and 受保护片段; the dichotomy-pattern list in both 风格目标 and references/machine-patterns.md) and could be tightened. | 2 / 3 |
Actionability | Gives concrete before→after rewrite recipes (e.g. '真实边界…弱化为本研究将 A 限定在…，重点处理 B'), exact verbatim-preserve token lists (\cite{}, \ref{}, \label{}, \eqref{}), and explicit parameter enums — copy-apply-ready guidance with no vagueness. | 3 / 3 |
Workflow Clarity | A clear 9-step workflow with explicit validation feedback loops (step 7 checks structure/verbatim preservation; step 8 runs 1-2 self-eval rounds with a specific check '是否误把更具体写成了新增事实'), appropriate for a zero-loss-sensitive editing task. | 3 / 3 |
Progressive Disclosure | Section-organized body with a single clearly signaled one-level-deep reference (references/machine-patterns.md, which exists); detailed pattern catalog is appropriately split out from the inline rules. | 3 / 3 |
Total | 11 / 12 Passed |