$npx -y skills add OpenRaiser/PaperFit --skill writing-polish本技能为 Semantic Polish Agent 提供具体、可执行的语义微调策略与禁区规则。它定义了在排版手段用尽后,如何通过最小化文字增删来消除孤行寡行、控制页数预算或优化末页留白,同时严格保持学术内容的原意、数据和结论不变。
| 1 | # Writing Polish Skill |
| 2 | |
| 3 | ## 概述 |
| 4 | |
| 5 | 本技能为 **Semantic Polish Agent** 提供具体、可执行的语义微调策略与禁区规则。它定义了在排版手段用尽后,如何通过最小化文字增删来消除孤行寡行、控制页数预算或优化末页留白,同时严格保持学术内容的原意、数据和结论不变。 |
| 6 | |
| 7 | 该技能不直接被 `orchestrator-agent` 调用,而是作为 `semantic-polish-agent` 的知识库和行为规范。所有语义级改写必须遵循本技能中定义的技巧和约束。 |
| 8 | |
| 9 | ## 适用场景 |
| 10 | |
| 11 | | 触发缺陷 | 操作方向 | 允许的改写幅度 | |
| 12 | |----------|----------|---------------| |
| 13 | | A1(孤行寡行) | 缩短 1-2 行 | 增删 3-8 词 | |
| 14 | | A2(末页留白) | 扩展 2-4 行 | 增加 20-50 词 | |
| 15 | | A3(页数预算) | 缩短或扩展多行 | 按需,但需分段执行 | |
| 16 | | 用户主动请求 | 精炼或扩写特定段落 | 用户指定 | |
| 17 | |
| 18 | ## 输入规范 |
| 19 | |
| 20 | | 输入项 | 来源 | 说明 | |
| 21 | |--------|------|------| |
| 22 | | 目标段落源码 | `semantic-polish-agent` 提取 | 需改写的一个或多个完整段落 | |
| 23 | | 改写目标 | 调用方请求 | `shorten` 或 `expand`,及期望行数变化 | |
| 24 | | 写作规范 | `config/writing_rules.yaml` | 时态、术语、禁用词等约束 | |
| 25 | | 上下文段落 | `semantic-polish-agent` 提取 | 前后各一段,用于保证语义连贯 | |
| 26 | |
| 27 | ## 输出规范 |
| 28 | |
| 29 | 本技能输出改写后的文本及变更元数据,供 `semantic-polish-agent` 整合为最终报告。 |
| 30 | |
| 31 | ```json |
| 32 | { |
| 33 | "skill": "writing-polish", |
| 34 | "changes": [ |
| 35 | { |
| 36 | "paragraph_id": 3, |
| 37 | "action": "shorten", |
| 38 | "net_word_change": -6, |
| 39 | "before_snippet": "It is worth noting that our method achieves state-of-the-art performance on several benchmark datasets.", |
| 40 | "after_snippet": "Our method achieves state-of-the-art results on several benchmarks.", |
| 41 | "techniques_used": ["remove_redundant", "phrase_to_word"], |
| 42 | "rationale": "移除冗余修饰词,将 'achieves state-of-the-art performance on several benchmark datasets' 压缩为 'achieves state-of-the-art results on several benchmarks'。语义等价,数据未变。" |
| 43 | } |
| 44 | ], |
| 45 | "warnings": [] |
| 46 | } |
| 47 | ``` |
| 48 | |
| 49 | ## 改写策略 |
| 50 | |
| 51 | ### 通用原则 |
| 52 | |
| 53 | 1. **最小修改优先**:能改一词不改一句,能改一句不改一段。 |
| 54 | 2. **保持学术严谨**:绝不改变数据值、引用标记、专有名词、核心声明。 |
| 55 | 3. **局部影响评估**:每次改写后需编译验证,确保不引入新的孤行或溢出。 |
| 56 | 4. **可逆性**:保留改写前文本,便于人工审查或回滚。 |
| 57 | |
| 58 | --- |
| 59 | |
| 60 | ### 策略组 1:缩短(Shorten) |
| 61 | |
| 62 | 目标:在不损失信息的前提下减少字数/行数。 |
| 63 | |
| 64 | #### 技巧 1.1:删除冗余修饰词 |
| 65 | |
| 66 | 移除对学术内容无实质贡献的修饰语。 |
| 67 | |
| 68 | - 删除强调性副词:`very`、`quite`、`extremely`、`highly` |
| 69 | - 删除填充短语:`It is worth noting that`、`It should be emphasized that`、`It is important to mention that` |
| 70 | - 删除冗余限定:`in a certain sense`、`to some extent` |
| 71 | |
| 72 | 示例: |
| 73 | |
| 74 | ``` |
| 75 | 修改前:It is worth noting that our method achieves very competitive performance. |
| 76 | 修改后:Our method achieves competitive performance. |
| 77 | 减少:5 词 |
| 78 | ``` |
| 79 | |
| 80 | #### 技巧 1.2:短语替换为单词 |
| 81 | |
| 82 | 用更简洁的单词或缩写替代多词短语。 |
| 83 | |
| 84 | | 原短语 | 替换为 | |
| 85 | |--------|--------| |
| 86 | | `in order to` | `to` | |
| 87 | | `a large number of` | `many` | |
| 88 | | `due to the fact that` | `because` | |
| 89 | | `at the present time` | `now` | |
| 90 | | `state-of-the-art methods` | `SOTA methods`(需已定义) | |
| 91 | | `with respect to` | `regarding` 或 `on` | |
| 92 | |
| 93 | 示例: |
| 94 | |
| 95 | ``` |
| 96 | 修改前:We conduct experiments in order to evaluate the performance of the proposed approach. |
| 97 | 修改后:We conduct experiments to evaluate our approach. |
| 98 | 减少:5 词 |
| 99 | ``` |
| 100 | |
| 101 | #### 技巧 1.3:被动语态转主动语态 |
| 102 | |
| 103 | 主动语态通常更简短且更有力。 |
| 104 | |
| 105 | ``` |
| 106 | 修改前:The experiments were conducted by us on three datasets. |
| 107 | 修改后:We conducted experiments on three datasets. |
| 108 | 减少:3 词 |
| 109 | ``` |
| 110 | |
| 111 | #### 技巧 1.4:合并相邻短句 |
| 112 | |
| 113 | 将两个紧密相关的短句合并为一句。 |
| 114 | |
| 115 | ``` |
| 116 | 修改前:We used the Adam optimizer. The learning rate was set to 1e-4. |
| 117 | 修改后:We used Adam with a learning rate of 1e-4. |
| 118 | 减少:6 词 |
| 119 | ``` |
| 120 | |
| 121 | #### 技巧 1.5:使用标准学术缩写 |
| 122 | |
| 123 | 在全文首次定义后,使用公认缩写。 |
| 124 | |
| 125 | | 原词 | 缩写 | |
| 126 | |------|------| |
| 127 | | `state-of-the-art` | `SOTA` | |
| 128 | | `natural language processing` | `NLP` | |
| 129 | | `mean average precision` | `mAP` | |
| 130 | |
| 131 | ``` |
| 132 | 修改前:Our method outperforms previous state-of-the-art approaches on the natural language processing benchmark. |
| 133 | 修改后:Our method outperforms previous SOTA approaches on the NLP benchmark. |
| 134 | 减少:5 词(假设 SOTA/NLP 已定义) |
| 135 | ``` |
| 136 | |
| 137 | #### 技巧 1.6:简化从句结构 |
| 138 | |
| 139 | 将定语从句压缩为分词短语或前置定语。 |
| 140 | |
| 141 | ``` |
| 142 | 修改前:The model which is trained on ImageNet achieves high accuracy. |
| 143 | 修改后:The ImageNet-trained model achieves high accuracy. |
| 144 | 减少:3 词 |
| 145 | ``` |
| 146 | |
| 147 | --- |
| 148 | |
| 149 | ### 策略组 2:扩展(Expand) |
| 150 | |
| 151 | 目标:在不注水的前提下增加有实质内容的文字。 |
| 152 | |
| 153 | #### 技巧 2.1:显式化隐含因果关系 |
| 154 | |
| 155 | 在结果陈述后补充简短的原因解释。 |
| 156 | |
| 157 | ``` |
| 158 | 修改前:Our method outperforms the baseline by 3.2%. |
| 159 | 修改后:Our method outperforms the baseline by 3.2%, likely because the attention mechanism better captures long-range dependencies. |
| 160 | 增加:11 词 |
| 161 | ``` |
| 162 | |
| 163 | #### 技巧 2.2:补充结果解释 |
| 164 | |
| 165 | 在表格或数据引用后,增加一句对关键发现的解读。 |
| 166 | |
| 167 | ``` |
| 168 | 修改前:Table 2 shows the ablation results. |
| 169 | 修改后:Table 2 summarizes the ablation study. Removing the temporal module causes a significant drop of 5.1%, confirming its importance for sequential modeling. |
| 170 | 增加:18 词 |
| 171 | ``` |
| 172 | |
| 173 | #### 技巧 2.3:强化与相关工作的对比 |
| 174 | |
| 175 | 在提及已有工作时,增加具体的差异说明。 |
| 176 | |
| 177 | ``` |
| 178 | 修改前:Unlike previous work, we use a transformer-based architecture. |
| 179 | 修改后:Unlike previous work that relied on recurrent networks with limited parallelization, we adopt a transformer architecture that scales more efficiently to long sequences. |
| 180 | 增加:14 词 |
| 181 | ``` |
| 182 | |
| 183 | #### 技巧 2.4:添加局限性讨论 |
| 184 | |
| 185 | 在结论或讨论部分,补充一句对当前方法局限性的客观陈述。 |
| 186 | |
| 187 | ``` |
| 188 | 修改前:Future work will explore larger-scale datasets. |
| 189 | 修改后:Future work will explore larger-scale datasets. A current limitation is the reliance on pre-trained word embeddings, which may not fully capture domain-specific terminology. |
| 190 | 增加:17 词 |
| 191 | ``` |
| 192 | |
| 193 | #### 技巧 2.5:拆分长句为短句 |
| 194 | |
| 195 | 通过增加句号拆分长句,可在不显著增加内容的情况下扩展行数。 |
| 196 | |
| 197 | ``` |
| 198 | 修改前:Our method consists of three components: an encoder, a decoder, and a refinement module. |
| 199 | 修改后:Our method consists of three components. First, the encoder extracts features from the input. Second, the decoder generates initial predictions. Finally, the refinement module iteratively improves the output. |
| 200 | 增加:14 词,行数增加更多 |
| 201 | ``` |
| 202 | |
| 203 | #### 技巧 2.6:补充技术细节(谨慎) |
| 204 | |
| 205 | 在不泄露未公开信息的前提下,可适当补充已在论文其他部分出现过的技术细节。 |
| 206 | |
| 207 | ``` |
| 208 | 修改前:We |