home / skills / huangwb8 / chineseresearchlatex / guide-updater

guide-updater skill

/skills/guide-updater

This skill updates project guidelines with conversational insights, inserting incremental bullets into the guide to ensure terminology consistency, stable

npx playbooks add skill huangwb8/chineseresearchlatex --skill guide-updater

Review the files below or copy the command above to add this skill to your agents.

Files (5)
SKILL.md
3.3 KB
---
name: guide-updater
description: 当用户明确要求"更新项目指南""同步指南""沉淀洞见到指南"时使用。将对话中新产生的可复用写作洞见实时沉淀到项目指南文件,保持术语口径一致、结构稳定、可检验与可复现。调用时必须指定指南文件路径。
metadata:
  author: Bensz Conan
  short-description: 项目指南实时更新器
  keywords:
    - 项目指南
    - 写作洞见
    - 知识沉淀
    - 术语一致
    - 指南更新
  triggers:
    - 更新项目指南
    - 同步指南
    - 沉淀到指南
    - guide updater
config: skills/guide-updater/config.yaml
---

# 项目指南更新器

将对话中新产生的可复用写作洞见实时沉淀到项目指南文件,保持术语口径一致、结构稳定、可检验与可复现。

## 使用方式

调用此技能时,**必须**在参数中指定指南文件路径:

```
/guide-updater --guide-path <路径>
```

或自然语言形式:
- "更新项目指南(路径:projects/MyProject/项目指南.md)"
- "把这些内容同步到指南文件 docs/写作指南.md"

## 先决条件

每次更新前快速自检:

- 不新增新的 `##` 级标题;只在既有小节/列表中"融入"新增要点。
- 不改动既有标题文本、编号、顺序;不为"整理美观"重排大量内容。
- 新增内容默认写成 1 条或少量 bullet,紧贴相邻条目的写法与语气。
- 涉及事实/文献:只写可核验信息;无法核验则标注"待核验",不要编造详细信息。

## 抽取"原子洞见"

- 将用户表达拆成若干条"可落笔"的要点,每条只包含一个主张/规则/写法。
- 为每条要点附 1 句"写进指南的目的"(例如:统一口径/防评审质疑/可复现/风险控制)。

## 选择融入位置

- 需要更细的落点规则时,先读 `skills/guide-updater/references/guide-structure-map.md` 再动手。
- 写作原则/禁区/句式 → `## 写作哲学`
- 术语、缩写、大小写、核心概念定义 → `## 命名与术语`
- 方法链条、验证口径、评估指标 → `## 基本原理`
- 范式定位、对照坐标系、差异化要点 → `## 文献综述与创新性`
- 具体写到哪一节(文档结构映射) → `## 落点清单`
- 评审问答口径 → `## 评审专家可能的问题及回答`
- 风险与对策 → `## 研究风险`
- "可核验事实锚点"(源码、参数、流程) → `## 附录`

不确定放哪:优先放到"最能减少未来写作返工"的位置;仍不确定就先问用户"更希望作为写作哲学/技术路线/风险控制哪一类沉淀?"再改文件。

## 以最小补丁写入

- 只做"增量补丁":新增 1--3 条 bullet 或在既有 bullet 末尾补 1 句澄清。
- 不新开大段长文;如果洞见很大,先拆成多个 bullet 分散落在对应小节。
- 与已有术语保持一致(参考项目既有术语表)。

## 更新后校验

- 运行 `python skills/guide-updater/scripts/validate_guide.py <指南文件路径>`(或手动确认关键 `##` 标题仍存在)。
- 搜索是否意外新增 `##` 级标题或引入"另起炉灶"的新结构。

## Resources

| 目录 | 用途 |
|------|------|
| `scripts/` | 校验 `项目指南.md` 是否保持关键结构与必要标题 |
| `references/` | 记录结构映射与落点规则,避免反复判断放置位置 |

Overview

This skill updates a project guide by persisting new, reusable writing insights from a conversation directly into the specified guide file. It aims to keep terminology consistent, structure stable, and documentation reproducible and verifiable. Calls must include the guide file path to apply changes.

How this skill works

On invocation with a guide path, the skill extracts atomic writing insights from the dialog and maps each into a single actionable bullet plus a one-sentence purpose. It locates the best insertion point using a guide-structure map and applies a minimal incremental patch (typically 1–3 bullets or a short clarifying sentence). After writing, it runs or recommends lightweight validation to ensure key headings and structure remain unchanged.

When to use it

  • When the user explicitly asks to “update project guide”, “sync guide”, or “persist insights to the guide”.
  • After a review discussion that produced reusable rules, terminology clarifications, or verification criteria.
  • When small, verifiable additions are needed without reshuffling existing guide structure.
  • When you need to enforce consistent terminology, evaluation rules, or reviewer Q&A don't appear in the guide.
  • When draft insights must be preserved immediately to avoid future rework.

Best practices

  • Always supply --guide-path (or a natural-language guide path) when calling the skill.
  • Prefer minimal patches: add 1–3 bullets or append one clarifying sentence to an existing bullet.
  • Do not create new top-level '##' headings; merge into existing sections that match the insight.
  • Attach a one-line purpose to each new bullet (e.g., unify wording, reduce review risk, ensure reproducibility).
  • Mark unverified factual claims as 'to be verified' rather than inventing details.
  • Run the provided validation script after updates to confirm the guide’s key headings remain intact.

Example use cases

  • A reviewer suggests a consistent phrasing for method descriptions—persist the phrasing and state the purpose as 'unify terminology'.
  • During discussion a new evaluation metric is agreed—add a bullet under 'Basic Principles' with verification criteria.
  • A lab member proposes a risk mitigation step—add it under 'Research Risks' with the purpose 'risk control'.
  • Clarify an abbreviation or casing rule by adding an entry to 'Naming and Terminology'.
  • Capture a candidate reviewer Q&A under 'Reviewer Questions and Answers' to prepare consistent responses.

FAQ

What if I don't know where to place an insight?

If placement is unclear, consult the guide-structure map; if still uncertain, ask the user whether it should be treated as philosophy, technical route, or risk control before writing.

Can the skill rewrite large sections for clarity?

No. The skill performs only incremental additions. Large reorganizations require explicit user approval and separate editing steps.