split-by-title-context

name	split-by-title-context
description	Split long text by headings and local context windows. / 按标题和上下文窗口切分长文。
metadata	{"builtin_skill_version":"0.1.0","copaw":{"emoji":"🧱","requires":{}}}

split-by-title-context

Purpose / 目标

Design a heading-aware segmentation result so each chunk remains semantically coherent for proofreading.

设计基于标题层级的切分方案，尽量保证每个分块语义完整，便于校对。

When to use / 何时使用

• "Split this long article before proofreading."
• "先把这本书按章节切块。"
• "Need context-preserving chunks for LLM review."

Inputs / 输入

• text: required, source content
• max_chunk_chars: optional, default 1500 to 2500
• min_chunk_chars: optional, default 500
• heading_levels: optional, for example H1-H3

Output / 输出

• chunk_plan: ordered chunk definitions
• context_links: previous and next context hints
• edge_cases: sections that need manual review

Boundaries / 边界

• Do not remove original content.
• Do not break inside code blocks, tables, or quotes when possible.
• Mark weak split points instead of forcing hard segmentation.

Execution Guidance / 执行指引

1. Identify heading structure and section boundaries.
2. Build candidate chunks under max length.
3. Merge very short neighboring chunks.
4. Add context notes for each chunk.

Prompt Contract / 提示约定

• Return a deterministic chunk index.
• Use explicit rationale for non-trivial merges.
• Surface uncertain boundaries.

Example I/O / 示例输入输出

Input example:

• task: "Apply this skill to the provided text block"
• constraints: "Keep original intent and only apply necessary changes"

输入示例：

• 任务："对给定文本应用本技能"
• 约束："保持原意，仅做必要修改"

Output example:

• result: concise, structured outcome
• highlights: key findings and priority actions
• follow_up: unresolved items for human review

输出示例：

• 结果：结构化且简洁的处理结果
• 重点：关键发现与优先处理项
• 后续：需人工确认的未决事项