refactor: reframe Step 5.3 from anti-detection to quality assessment
- Rename "Tier 3 评估" to "综合评估", describe dimensions directly (tone variance, density rhythm, pacing, readability) without referencing anti-detection framework - Reframe composite_score from "0=human, 100=AI" to "0=high quality, 100=issues found" - Change 5.3 role from "gate control" to "supplementary verification" Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
parent
7988bb4e99
commit
96f6df6209
1 changed files with 4 additions and 4 deletions
8
SKILL.md
8
SKILL.md
|
|
@ -352,15 +352,15 @@ LLM 自行完成,不需要调用脚本。
|
|||
|
||||
不通过 → **定向修复**:只替换不达标的具体句子/段落,不动已通过的部分。每轮最多改 3 处,改完立即重新检查该项。2 轮仍不过 → 标注跳过,继续下一项。
|
||||
|
||||
**5.3 脚本验证**(补充逐项检查):
|
||||
**5.3 脚本辅助验证**(补充 5.2 的逐项检查):
|
||||
|
||||
Agent 在 5.2 逐项检查时同步完成 Tier 3 评估(风格漂移、密度波浪、连贯性打破、整体人感),产出 0-1 分数。
|
||||
Agent 在 5.2 检查过程中同步完成综合评估(各 H2 之间的语气差异度、信息密度的高低交替、段落间的节奏变化、整体阅读流畅度),产出 0-1 分数。
|
||||
|
||||
```bash
|
||||
python3 {skill_dir}/scripts/humanness_score.py {article_path} --json --tier3 {agent_tier3_score}
|
||||
```
|
||||
|
||||
解读 JSON 中 `composite_score`:
|
||||
解读 JSON 中 `composite_score`(0=质量高, 100=问题多):
|
||||
- < 30 → 通过,继续 Step 6
|
||||
- 30-50 → 查看 `param_scores` 中最低分的 1-2 项,只修复对应的具体句子(不重写整段),改完重新打分。1 轮即可
|
||||
- \> 50 → 取 `param_scores` 最低的 2-3 项,逐项定向修复(每项只改最相关的 1-2 处),最多 2 轮。仍 > 50 则标记 DONE_WITH_CONCERNS 继续
|
||||
|
|
@ -435,7 +435,7 @@ python3 {skill_dir}/toolkit/cli.py preview {markdown} --theme {theme} --no-open
|
|||
dimensions:
|
||||
- "{维度}: {选项}"
|
||||
closing_type: "{收尾类型}" # trailing_off/unanswered/scene_revert/abrupt_stop/anti_conclusion/image
|
||||
composite_score: {Step 5.3 的 composite_score} # 0=人类, 100=AI
|
||||
composite_score: {Step 5.3 的 composite_score} # 0=质量高, 100=问题多
|
||||
writing_config_snapshot: # 本次使用的关键参数(从 writing-config.yaml 提取)
|
||||
sentence_variance: {值}
|
||||
paragraph_rhythm: "{值}"
|
||||
|
|
|
|||
Loading…
Reference in a new issue