Back to Details

refine-file

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Refine File

精炼文件

Apply the requested subset of WORTH → PLACE → SHAPE to one file: S, W+S, or W+P+S.
针对单个文件应用WORTH（价值评估）→ PLACE（位置评估）→ SHAPE（结构优化）流程的指定子集：仅S、W+S，或W+P+S。

Steps

步骤

Step 0 — Resolve the file and lens subset

步骤0 — 确定文件与流程子集

Read the named file. This workflow applies WORTH and PLACE only to current guidance. An active-state document or dated investigation can receive S here; use 
place-fact
and its owning workflow for separate placement and lifecycle decisions.
Choose the smallest subset that answers the request:
  • S — the user explicitly invoked this skill for shape-only work.
  • W+S — the user asks what is worth keeping or requests pruning without a placement audit.
  • W+P+S — the user asks to refine, audit, re-home, or judge whether material belongs in the file.
State the subset and proceed when the request selects it. Ask only when two subsets would materially change the result.
The named file is the audit scope, not the edit boundary; MOVE may edit its canonical target and required delivery views.
读取指定文件。本工作流仅对当前指导内容应用WORTH和PLACE评估。对于活跃状态的文档或有日期标记的调研文档，可在此仅应用S流程；若需单独进行位置和生命周期决策，请使用
place-fact
及其所属工作流。
选择能满足请求的最小流程子集：
  • S — 用户明确要求仅进行结构优化工作。
  • W+S — 用户询问哪些内容值得保留，或要求精简但无需进行位置合理性审核。
  • W+P+S — 用户要求优化、审核、重新归置内容，或判断材料是否属于该文件。
当请求明确指定子集时，说明所选子集并继续执行。仅当两个子集会对结果产生实质性影响时，才需向用户确认。
指定文件为审核范围，而非编辑边界；MOVE操作可能会修改其标准目标文件及所需的交付视图。

Step 1 — Load the selected lenses

步骤1 — 加载选定的流程工具

Invoke only the selected skills via the Skill tool:
SubsetSkills
S
tighten-instruction
, 
structure-prose
W+S
vet-fact
, 
tighten-instruction
, 
structure-prose
W+P+S
vet-fact
, 
place-fact
, 
tighten-instruction
, 
structure-prose
This file-level workflow owns the apply gate. The lenses never invoke each other.
通过Skill工具仅调用选定的技能：
子集技能
S
tighten-instruction
, 
structure-prose
W+S
vet-fact
, 
tighten-instruction
, 
structure-prose
W+P+S
vet-fact
, 
place-fact
, 
tighten-instruction
, 
structure-prose
本文件级工作流负责管控应用权限，各流程工具之间不得互相调用。

Step 2 — Find and score independent edits

步骤2 — 找出并评估独立编辑项

Apply the selected lenses per fact in WORTH → PLACE → SHAPE order:
  • CUT — the fact fails 
    vet-fact
    ; record the current text and a one-line reason.
  • MOVE — 
    place-fact
    keeps the fact but selects another owner or delivery boundary; record the fact, WORTH category, and target.
  • SHAPE — the kept, in-place fact or block needs tightening or structure; record current → proposed and the level: whole file, section/block, or line.
A WORTH cut dissolves PLACE and SHAPE work for the same fact. Preserve rationale when its reason is the non-derivable constraint; shape it as 
behaviour — constraint
.
For each MOVE:
  1. Record 
    place-fact
    's full placement contract in the analysis; do not paste it into the repository.
  2. Inspect the existing target, or confirm the contract for a new artifact, before scoring.
  3. Shape the fact for the canonical target, establish any required derived or guarded delivery view, then remove it from the source.
  4. If the target already carries the fact, skip the add and classify the source removal as CUT.
Score confidence 
0.00–1.00
that the fact assessment, action, and resulting text are correct. A MOVE score also covers the target and delivery contract.
按照WORTH → PLACE → SHAPE的顺序，对每个事实应用选定的流程工具：
  • CUT（删除） — 事实未通过
    vet-fact
    验证；记录当前文本及一行理由。
  • MOVE（迁移） — 
    place-fact
    判定事实需保留，但需归至其他所有者或交付边界；记录该事实、WORTH分类及目标位置。
  • SHAPE（优化） — 保留且位置合理的事实或内容块需要调整格式或结构；记录当前内容→修改后内容，以及优化级别：整个文件、章节/内容块，或单行文本。
若某事实因WORTH评估被删除，则无需再对其进行PLACE和SHAPE操作。当理由为非推导性约束时，需保留该依据，并按
行为 — 约束
的格式整理。
对于每个MOVE操作：
  1. 在分析记录中完整记录
    place-fact
    的位置协议，请勿将其粘贴至代码仓库。
  2. 在评估前，检查现有目标文件，或确认新工件的协议。
  3. 针对标准目标文件优化该事实内容，建立所需的衍生或受保护交付视图，然后从源文件中移除该内容。
  4. 若目标文件已包含该事实，则跳过添加操作，并将源文件中的移除操作归类为CUT。
为事实评估、操作及最终文本的正确性打分（0.00–1.00）。MOVE操作的评分还需涵盖目标位置及交付协议的合理性。

Step 3 — Gate and apply

步骤3 — 审核并应用编辑

Show the ordered plan, then apply every finding at 
c ≥ 0.75
and hold lower-confidence findings without changing their text. Apply CUT → MOVE → SHAPE; within SHAPE, apply whole-file → section/block → line. Drop any queued finding that an earlier edit dissolves.
text
**Refinement plan — <file>:**
- [0.93] CUT: <current> — <reason>
- [0.86] MOVE: <fact> → <canonical target>
- [0.78] SHAPE: <current> → <proposed>
- [0.64] CUT: <current> — <reason> · held (< 0.75)
Write 
Nothing to refine — every fact is worth keeping, well placed, and clear.
when there are no findings.
展示排序后的优化计划，然后应用所有置信度≥0.75的结论，置信度较低的结论暂不修改原文。按CUT → MOVE → SHAPE的顺序应用；在SHAPE操作中，按整个文件→章节/内容块→单行文本的顺序执行。若某编辑项因之前的操作而失效，则丢弃该队列中的编辑项。
text
**优化计划 — <文件>:**
- [0.93] CUT: <当前内容> — <理由>
- [0.86] MOVE: <事实内容> → <标准目标文件>
- [0.78] SHAPE: <当前内容> → <修改后内容>
- [0.64] CUT: <当前内容> — <理由> · 暂不执行（置信度<0.75）
若无任何优化结论，需注明：“无需优化 — 所有事实均值得保留、位置合理且表述清晰。”

Step 4 — Prove the result cold

步骤4 — 验证最终结果

Re-read the named file and every MOVE target in full. Confirm every kept fact and load-bearing rationale survived, every edit stayed inside the selected subset, and the result has no contradiction, duplicate ownership, mixed delivery scope, broken route, or unnecessary eager loading. Verify that each moved fact has one canonical owner and that every required delivery view derives from it or has an automatic equality guard.
Revert or fix any edit that fails this proof.
完整重读指定文件及所有MOVE操作的目标文件。确认所有保留的事实及关键依据均未丢失，所有编辑操作均未超出选定的流程子集范围，且结果无矛盾、无重复归属、无混合交付范围、无路径断裂或不必要的预加载问题。验证每个迁移的事实仅有一个标准所有者，且所有所需的交付视图均源自该所有者，或具备自动一致性保障。
若验证不通过，需回滚或修正相关编辑操作。

Step 5 — Report

步骤5 — 生成报告

text
**Refinement result — <file>:**
- Applied: [N cut, N moved, N shaped | none]
- Held: [finding, score, and reason, one per line | none]
- Dissolved: [finding removed by an earlier edit, one per line | none]
- Delta: [net lines and words removed]
text
**优化结果 — <文件>:**
- 已应用：[N项删除，N项迁移，N项优化 | 无]
- 暂不执行：[结论、评分及理由，每行一项 | 无]
- 已失效：[因之前编辑操作而被移除的结论，每行一项 | 无]
- 变化：[净删除的行数和字数]