跳转到主要内容
技能是赋予 Agent 深度、任务专属知识的 Markdown 文件。为短篇小说创作定义一个,为拆书笔记定义一个,为剧本格式化定义一个——Agent 只加载与当前任务相关的技能。

技能的工作原理

技能使用渐进式加载来保持上下文精简。打开项目时,Agent 只看到所有技能的名称和描述——每个技能只有几十个字。这是启动时唯一加载的内容。 当你的请求与某个技能的描述匹配时,Agent 激活该技能并加载完整指令。短篇小说创作技能在你写查询信时保持休眠,恐怖小说技能在你做惊悚小说大纲时不会占用上下文。
级别加载时机内容
元数据始终,启动时frontmatter 中的名称和描述
指令技能触发时完整的 SKILL.md 正文
支撑文件按需模板、脚本、参考文档

技能结构与存储位置

每个技能是一个包含 SKILL.md 文件的目录,目录名即为技能名。技能还可以在三个可选子目录中包含支撑文件。
your-novel/
├── .soloent/
│   └── skills/                  # 项目技能(仅限本项目)
│       └── short-story/
│           ├── SKILL.md                   # 必须:名称、描述、指令
│           ├── templates/                 # 可选:输出模板、格式指南
│           │   ├── chapter-outline.md
│           │   └── character-profile.md
│           ├── docs/                      # 可选:参考资料、查找表
│           │   └── genre-conventions.md
│           └── scripts/                   # 可选:Agent 可运行的处理脚本
│               └── word-count.py
├── SOLOENT.md
├── chapters/
└── ...
支撑文件只在指令引用时才加载,启动时不占用上下文。模板在 Agent 执行到需要它的步骤之前保持休眠,参考文档只在 Agent 需要查阅时才被读取。 全局技能存储在系统级技能目录中,在所有项目中都可以使用。适合跨项目通用的能力,例如生成开篇钩子的技能,或引导 Agent 完成标准作品审核流程的技能。 项目技能存放在项目根目录的 .soloent/skills/ 中。适合与特定项目或类型绑定的能力——基于这个言情系列特定规则构建的技能,或针对当前故事的尺度和叙事风格校准的技能。
全局技能与项目技能同名时,项目技能优先。

技能开关

每个技能都有开关,无需删除文件即可启用或禁用。当前任务不需要某个技能时,禁用它可以让它脱离上下文,避免意外激活。

创建技能

SKILL.md 文件由两部分组成:frontmatter(元数据)和指令正文。
---
name: my-skill
description: 描述何时触发此技能,而不是它能做什么。
---

技能激活时 Agent 应遵循的指令。
两个必填 frontmatter 字段:
  • name 必须与目录名完全一致
  • description 描述何时触发此技能,而不是它能做什么(不超过 1,024 个字符)
编辑器中打开的 SKILL.md 文件,显示 name 和 description 字段,右侧 Skills 面板将 english-short-story 技能列于工作区技能下
1

打开 Skills 菜单

点击 Agent 面板左数第四个图标,切换到 Skills 管理器。
2

新建技能

点击”New skill…”,输入名称(例如 short-story),再点击”+“。Agent 会创建包含模板 SKILL.md 文件的技能目录。
3

编写技能指令

编辑 SKILL.md 文件:
  • 更新 description,指定此技能的触发时机
  • 在正文中添加指令、检查清单和工作流步骤
  • 可选择在 templates/docs/ 子目录中添加支撑文件
以下是官方提供的 english-short-story 完整技能文件——这是一个领域专家型技能,包含完整的 6 步工作流、检查清单和停止条件,可以直接复制使用。
---
name: english-short-story
description: A SoloEnt official writing skill. Create English short fiction with strong hooks, sharp reversals, and high emotional intensity using a 6-step workflow: concept brainstorm, character design, outline, chapter writing with auto-review, and optional chapter review and title/hook generation. Explicit scenes and dark emotional registers are available tools, not requirements. Use when writing short stories, drafting chapters, designing characters, building outlines, reviewing fiction chapters, or generating titles and hooks.
---

# English Short Story

Manages the full creation workflow for English short fiction (10–20 chapters, 900–1,200 words each). Designed for stories built on strong hooks, sharp reversals, and high emotional intensity — across any register including thriller, horror, romance, erotic, literary, or hybrid. Explicit scenes and dark emotional content are available tools calibrated per story, not default requirements.

---

## Before Any Task: Read SOLOENT.md

If `SOLOENT.md` exists, read it and extract:

- **Section 8.1 Todo List** — what is the current pending task?
- **Section 7.1 Current Status** — which step/chapter are we on?
- **Section 7.4 Pending Threads** — active hooks and next chapter goal
- **Section 3 Character Index** — is the Character Voice Profile available?

If `SOLOENT.md` does not exist, proceed directly to Step 1. After the user confirms the core concept, run `/init` to create and populate `SOLOENT.md` before Step 2 begins.

---

## Usage

Use this skill whenever you are working on an English short fiction project that prioritizes strong hooks, emotional intensity, and sharp plot reversals. It manages the full creation process from a one-line premise through to finished chapter files. Works across all registers — thriller, horror, romance, erotic, literary, or hybrid. Explicit content level is calibrated per story during Step 1.

**Start a new story**: provide a one-line premise and the skill walks you through Steps 1–4 in sequence, requiring your confirmation at each stage before advancing. Steps 5 and 6 are optional and run on request.

Trigger phrases that activate this flow:
- Direct story ideas: "A woman finds a letter in her dead husband's coat…", "Two strangers trapped in a lighthouse…"
- Intent statements: "I want to write a short story", "I have a story idea", "Let's write a short fiction"
- Explicit skill invocation: "Use the short story skill", "Start the writing workflow", "短篇技能"

**Resume an existing story**: open a project that already has a `SOLOENT.md` and say "continue the story" or "pick up where we left off" — the skill reads current state and resumes from the correct step.

**Jump to a specific step**: name the step directly (e.g. "run Step 4 for Chapter 3", "review Chapter 5") and the skill starts there.

**Prompt templates are optional**: if `templates/Prompt-0X-*.md` files are present the skill loads them; if not, it runs on the built-in instructions and checklists.

---

## The 6-Step Workflow

### Step 1 — Core Concept Brainstorm

**Input**: One-line premise from user
**Output**: Core concept + story arc (3+ reversals + explicit calibration + sensory register)

If `templates/Prompt-01-Core Concept Brainstorm.md` exists, load and follow it. Otherwise proceed using the output spec and checklist below.

**Checklist before advancing to Step 2**:
- [ ] Central tension defined (desire / fear / obsession / transgression)?
- [ ] 3+ major plot reversals designed?
- [ ] Erotic / horror / thriller elements are structural, not decorative?
- [ ] Explicit level calibrated and noted?
- [ ] First-person narration viable?

**Wait for user confirmation.**

→ Once confirmed, run `/init` to initialize `SOLOENT.md` before proceeding to Step 2.

---

## /init — Initialize SOLOENT.md

Run automatically after Step 1 is confirmed, before Step 2 begins.

**Actions**:
1. Create `SOLOENT.md` in the project root.
2. Fill in the following fields from the Step 1 output:
   - **Section 1 Project DNA**: genre, core themes, emotional promise
   - **Section 5 Stylistic Guidelines**: base tone, language register, pacing notes derived from concept
   - **Section 7.1 Current Status**: story title, story time, immediate atmosphere
   - **Section 8.1 Todo List**: check off `[x] Confirm core concept (Step 1)`, leave remaining todos open
   - **Section 8.2 Milestone Log**: add `[Completed] Step 1 — core concept confirmed: [premise summary] / explicit level: [X]`
3. Confirm to user: `SOLOENT.md initialized — ready for Step 2.`

---

### Step 2 — Character Design

**Input**: Confirmed core concept
**Output**: Main cast + Character Voice Profile

If `templates/Prompt-02-Character Design.md` exists, load and follow it. Otherwise proceed using the output spec and checklist below.

**Rules**:
- Complete and confirm characters **before** touching the outline
- Must output a **Character Voice Profile**: 1–2 sentences per character (speech patterns + signature phrases) — pasted into every Step 4 prompt

**Checklist before advancing to Step 3**:
- [ ] Protagonist is active? Clear goal driving them forward?
- [ ] Every character embodies a value, desire, fear, or moral position?
- [ ] Protagonist's voice is distinct, layered, and evolves?
- [ ] Character Voice Profile output and ready for Step 4?
- [ ] Core cast ≤ 4 characters?
- [ ] Relationships generate natural conflict?

**Wait for user confirmation.**

→ Update `SOLOENT.md`:
- **Section 3.1 Character Index**: add each character entry with name, age, role, and pointer to `characters/[name].md`
- **Section 3.2 Relationship Network**: fill in key relationships
- **Section 7.2 Character Snapshot**: fill in current character states
- **Section 8.1**: check off character design todo, add `[ ] Complete story outline`
- **Section 8.2**: log `[Completed] Step 2 — characters confirmed: [names] / Voice Profile ready`

---

### Step 3 — Outline and Chapter Plan

**Input**: Confirmed concept + confirmed characters + chapter count (default 10–20)
**Output**: 3-act outline + chapter-by-chapter plan

If `templates/Prompt-03-Outline and Chapter Plan.md` exists, load and follow it. Otherwise proceed using the output spec and checklist below.

**Chapter plan format** (must match Step 4 input):
- Story beats: 3–5 per chapter
- Intensity peak: note trigger method (dialogue / action / discovery)
- Closing hook: note tension type (reversal / revelation / threat / disappearance)
- No pre-set timestamp transitions

**Checklist before advancing to Step 4**:
- [ ] Chapter count confirmed with user?
- [ ] Chapter 1 opens on immediate conflict?
- [ ] Chapter 1 intensity peak strong enough to lock the reader?
- [ ] Every chapter has: beats + intensity peak (trigger noted) + closing hook (tension type noted)?
- [ ] Reversal points clearly marked?
- [ ] Explicit / horror / thriller peaks distributed throughout, not clustered?

**Wait for user confirmation.**

→ Update `SOLOENT.md`:
- **Section 4.1 Story Structure**: fill in chapter range and act breakdown
- **Section 4.2 Outline Index**: add pointer `[Story Outline] -> outline/master-outline.md`
- **Section 8.1**: check off outline todo, add `[ ] Begin writing Chapter 1`
- **Section 8.2**: log `[Completed] Step 3 — outline confirmed: [X] chapters / [act structure summary]`

---

### Step 4 — Chapter Writing + Auto-Review

**Input**: Current chapter plan + Character Voice Profile; from Chapter 2 onward, also paste previous 1–2 chapters
**Output**: Chapter prose (900–1,200 words) + auto-review report

If `templates/Prompt-04-Chapter Writing.md` exists, load and follow it. Otherwise proceed using the output spec and checklist below.

**Per-chapter process**:
1. Chapter 1: paste chapter plan + Voice Profile
2. Chapter 2+: paste chapter plan + Voice Profile + previous 1–2 chapters
3. Run `wc -w chapters/ch-XX.md` for word count — do not use model self-estimate
4. Output auto-review immediately after prose

**Auto-Review Checklist** (run after every chapter, mark ✅/❌):
- [ ] Word count 900–1,200 (verified via `wc -w`)?
- [ ] All beats executed in order? Intensity peak landed with correct trigger?
- [ ] Closing hook sharp? No moral summary / reflection / epiphany ending?
- [ ] First-person POV clean — no "he thought / she felt" violations?
- [ ] No 3+ consecutive introspective sentences?
- [ ] Dialogue drives plot (≥ 50%)?
- [ ] Each character's speech matches their Voice Profile?
- [ ] No repeated phrasing within the same character's lines?
- [ ] No timestamp scene openers ("The next day…" / "Three weeks later…")?
- [ ] Age/occupation/backstory communicated obliquely after first mention?
- [ ] Explicit intensity consistent with Step 1 calibration?
- [ ] No AI tells (mechanical structures, em-dash overuse, clichéd phrasing)?

→ After each chapter, update `SOLOENT.md`:
- **Section 7.1**: update `Current Chapter` to next chapter number
- **Section 7.3 Recent Plot Flow**: replace with 3–5 key events from this chapter
- **Section 7.4 Pending Threads**: update `Urgent Hooks`, `Active Foreshadowing`, `Next Chapter Goal`
- **Section 7.5 Execution Review**: paste auto-review checklist result + word count
- **Section 8.1**: check off current chapter todo, add next chapter todo
- **Section 8.2**: log `[Completed] Ch-[X] written — [one-line plot summary]`

> For deep-dive review of specific chapters, use Step 5 (optional).

---

### Step 5 — Chapter Review (Optional)

**Input**: User-specified chapter(s)
**Output**: Detailed review report + revision recommendations

If `templates/Prompt-05-Chapter Review.md` exists, load and follow it. Otherwise proceed using the checklist below.

**When to trigger**:
- User questions a chapter's quality
- Reversal chapters need tonal polish
- Multi-chapter continuity check needed
- Explicit content consistency review across chapters

**Checklist**:
- [ ] Word count (900–1,200) + outline fidelity (beats + intensity peak)?
- [ ] Narrative logic continuity, prose style consistency, character voice fidelity?
- [ ] ⚠️ Tag repetition: age/occupation ≤ 2 mentions; timestamp openers = 0; no phrase repetition?
- [ ] Explicit escalation correct? Any tonal inconsistencies?
- [ ] No AI tells (mechanical structures, digressive monologues, em-dash overuse)?

→ Update `SOLOENT.md`:
- **Section 7.5 Execution Review**: log review findings and revision status for reviewed chapter(s)
- **Section 8.1**: check off review todo if applicable

---

### Step 6 — Title and Opening Hook (Optional)

**Input**: Completed full story (or concept + key scene summary)
**Output**: 100–150 word opening hook + 3 title options

If `templates/Prompt-06-Title and Hook.md` exists, load and follow it. Otherwise proceed using the output spec and checklist below.

**Run only when user requests.** Recommended after all chapters are complete — extract tension from the actual draft.

**Checklist**:
- [ ] Opening hook is 100–150 words, names core conflict, creates immediate suspense?
- [ ] 3 titles differ in style (conflict angle / intrigue angle / atmosphere angle)?
- [ ] Titles calibrated for English-language fiction platforms?

**Delivery**: individual chapter files + title and hook; merge into single file only if user requests.

→ Update `SOLOENT.md`:
- **Section 7.1**: set `Current Chapter` to `Complete`
- **Section 8.1**: check off Step 6 todo if present
- **Section 8.2**: log `[Completed] Step 6 — title and hook delivered / project complete`

---

## Core Creative Principles

- **Hook before everything**: the opening must create immediate tension — a question, a threat, a desire the reader cannot ignore
- **Emotional intensity over social melodrama**: drive the story with visceral feeling — desire, dread, obsession, grief, rage — not domestic drama for its own sake
- **Escalating stakes**: each act pushes further than the reader expects — in intimacy, danger, revelation, or moral cost
- **Active protagonist**: story engine is what the protagonist *chooses to do*, not what happens to them
- **Characters embody value**: each character represents a desire, a fear, a moral position, a kind of power
- **3+ major reversals** distributed across the full story
- **First-person narration** for maximum immersion
- **Explicit content is a calibrated tool**: level set in Step 1 — can range from zero to high depending on the story's needs

三种常见技能类型

类型一:检查清单型

适合在固定节点一次性核验多项标准,防止漏项。例如每章写完后的提交前检查:
---
name: chapter-submission-check
description: Use before finalizing a chapter to verify word count, consistency, and all checklist items.
---

## 提交前检查(全部通过才能推进)
- [ ] 字数 900–1,200(已用工具核实,非估算)
- [ ] 所有章节节拍已执行,强度峰值已落地
- [ ] 结尾钩子清晰,无道德总结 / 感悟收尾
- [ ] 人物名字一致性检查通过
- [ ] SOLOENT.md 已更新

## 输出
每项通过 / 未通过。任何未通过项必须修复后才能继续。

类型二:工作流型

适合有明确步骤顺序、执行前需要确认的操作。
---
name: chapter-rewrite
description: Rewrite a chapter from scratch. Run only when explicitly requested.
---

## 步骤
1. 备份:将当前章节另存为 ch-XX-original.md,保留原稿
2. 说明:用一段话总结本次重写的核心调整方向,等待用户确认后再动笔
3. 执行:按新方向写完整章节(900–1,200 字)
4. 验证:对照原版运行章节检查清单,输出新旧版本的主要差异

类型三:领域专家型

适合让 Agent 按专家经验工作、而不是凭感觉乱猜的情况。例如写作卡壳时的诊断技能:
---
name: story-diagnosis
description: Use when a chapter feels flat, pacing is off, or a scene isn't landing.
---

## 证据收集
1. 读取问题章节全文
2. 对照 SOLOENT.md 第 4 节核查章节节拍计划
3. 检查上下两章的情绪曲线

## 决策矩阵
| 症状 | 先查 |
|---|---|
| 节奏拖沓 | 节拍密度 / 有无无效段落 |
| 情感不到位 | 人物声音一致性 / 内心独白是否过多 |
| 结尾钩子失效 | 悬念类型 / 张力是否已在章内释放 |

## 输出格式
根本原因 / 影响范围 / 修复步骤 / 验证方式

完整案例:作品总结

这是一个领域专家型技能。 它定义了获取内容路径(脚本按需提取章节)、专业定义(docs/ 中存放章节识别模式)和固定输出格式(templates/ 中的笔记模板)——这三个要素正是领域专家型技能的典型结构。 此技能从本地书籍文件中提取结构化笔记——章节摘要、关键观点、金句——并保存为格式化 Markdown。

目录结构

book-summary/
├── SKILL.md                    # 工作流指令
├── templates/
│   └── note-template.md        # 读书笔记的输出格式
├── docs/
│   └── chapter-patterns.md     # 中英文书籍的章节标题识别模式
└── scripts/
    └── analyze-book.py         # 在 Agent 读取前预处理书籍文件

为什么需要脚本

书籍很长。把整本书加载到上下文中来提取章节结构既浪费又缓慢。预处理脚本解决了这个问题:
  • info 命令 — 读取前 50 行,识别书名、编码和总行数
  • chapters 命令 — 扫描章节标题并返回行号,无需读取全文
  • extract 命令 — 只读取某一章的内容,让 Agent 逐章处理
Agent 在工作流的每个阶段调用脚本。只有脚本的输出进入上下文,书籍本身不会。

为什么需要 docs 文件

中英文书籍的章节标题格式差异很大。与其把所有识别模式嵌入 SKILL.md,不如将它们存放在 docs/chapter-patterns.md 中。Agent 需要识别章节边界时读取这个文件,其他时候忽略它。 将参考资料放在 docs/ 中有两个好处:让 SKILL.md 专注于工作流逻辑,同时可以更新模式而不修改主指令。

为什么需要模板

输出格式是固定的:每份读书笔记都遵循相同结构——书籍概览、章节摘要(含关键观点和金句)、全书精华、个人思考。与其在每次提示词中重新说明格式,不如将其存放在 templates/note-template.md 中。 第 5 步(最终汇总)运行时,Agent 加载模板并填入占位符。格式每次一致,只需在一处维护。

SKILL.md 如何引用支撑文件

SKILL.md 中的指令明确指向每个支撑文件:
### 步骤 1:分析书籍结构

运行预处理脚本获取书籍信息:
  python3 scripts/analyze-book.py <文件路径> info

### 步骤 5:汇总生成笔记

按模板输出:templates/note-template.md

章节标题识别模式参考:docs/chapter-patterns.md

写出有效的技能

description 写”何时用我”,不写”我是干什么的”

这两个听起来差不多,但对触发准确率影响很大。
# 错误——描述的是功能,不是触发条件
description: This skill helps improve writing quality and checks chapters for common issues.

# 正确——描述的是何时该激活
description: Use when reviewing a completed chapter draft for pacing, voice, and consistency issues.
描述符常驻上下文,每个启用的技能都在持续消耗 token 配额。写短,写准,直接说触发场景:
# 低效(约 45 tokens)
description: |
  This skill helps you review and improve your writing chapters.
  It checks for common issues like pacing problems, character voice...
  Use this when you want to ensure your chapter quality before proceeding.

# 高效(约 9 tokens)
description: Use when reviewing a chapter for quality issues.

有完整步骤、输入、输出和停止条件

一个技能指令如果只写了流程开头、没有明确的完成标志,Agent 会自行决定什么时候停下来——通常不是你期望的位置。每个步骤都应该有:
  • 输入:这一步需要哪些上下文或文件
  • 输出:这一步产出什么
  • 停止条件:什么情况下等待确认,什么情况下继续

正文只放导航和核心约束,大资料拆到支撑文件

SKILL.md 正文控制在 5,000 字以内。如果需要风格参考、识别模式、角色档案等大量资料,放进 docs/ 子目录,在指令里按文件名引用——Agent 需要时才加载,不需要时不占上下文。

按调用频率决定是否开启自动触发

使用频率建议策略
高频(每次会话超过 1 次)保持自动触发,优化描述符让触发更准确
低频(每次会话不超过 1 次)关闭自动触发,手动调用,描述符可以脱离上下文
极低频(每月不超过 1 次)移除技能,改为在 SOLOENT.md 中记录文档

容易踩的坑

问题症状怎么修
描述过于宽泛description: help with writing——任何写作请求都能触发收窄到具体场景,加入实际触发短语
正文过长几百行工作手册全塞进 SKILL.md 正文参考资料拆到 docs/,指令只保留骨架
一个技能做五件事同一个技能覆盖审稿、续写、大纲、角色设计、诊断拆开,每个领域一个技能,可以单独开关

使用他人的技能

不必从头创作每一个技能。技能是自包含的目录——你可以手动安装、从仓库克隆,或以插件形式安装。

手动安装

下载或复制技能目录,放置到对应位置:
  • 全局技能 → 系统级技能目录(在所有项目中可用)
  • 项目技能 → 项目内的 .soloent/skills/
SoloEnt 检测到目录后技能立即激活,无需重启。
your-novel/
├── .soloent/
│   └── skills/
│       └── book-summary/      ← 将技能目录放在这里
│           ├── SKILL.md
│           ├── templates/
│           ├── docs/
│           └── scripts/

通过 git clone

托管在 GitHub 上的公开技能可以直接克隆到技能目录:
cd your-novel/.soloent/skills/
git clone https://github.com/author/skill-name
这也方便技能的后续更新——随时用 git pull 拉取最新改动。
如果希望技能在所有项目中都可用,克隆到全局技能目录而不是项目目录。

通过插件

部分技能作者以打包好的 .vsix 扩展文件分发技能。安装扩展后,内置技能自动加载,无需手动放置文件。
1

获取 .vsix 文件

从技能作者的仓库或分发渠道下载 .vsix 文件。
2

安装扩展

在 SoloEnt 中打开扩展面板,点击 ··· 菜单,选择从 VSIX 安装…,选中下载的文件。
3

确认技能已加载

打开 Skills 面板,扩展中内置的技能会出现在技能列表中,可以直接使用。
激活技能前请先查阅其 SKILL.md。技能在你的权限下运行,可以读取和修改你的项目文件。

故障排除

最可能的原因:描述不够具体,或者技能被关闭了。
  • 检查技能在 Skills 面板中是否已启用
  • 重写 description,加入你实际使用的短语和任务类型——“帮助写作”很少能正确触发;列出具体工作流、输出类型和示例触发语句
  • 确认技能目录中有有效的 SKILL.md 文件,且包含 namedescription 两个 frontmatter 字段
  • 确认 frontmatter 中的 name 与目录名完全一致
描述太宽泛了——它匹配了你并不想触发它的请求。
  • description 收窄,只描述这个技能处理的具体任务
  • 不需要某个技能时,在 Skills 面板中关掉它
  • 如果多个技能的描述有重叠,分别把每个描述写得更加具体,明确各自的适用范围
请检查以下常见问题:
  • 指令可能超出了上下文限制——把 SKILL.md 控制在 5,000 字以内,将参考资料移到 docs/ 子目录
  • 某条全局或项目规则可能覆盖了技能的部分行为——检查当前启用的规则是否与技能指令冲突
  • 如果技能使用了检查清单,确保每个关键节点都有明确的”等待确认”门控;否则 Agent 会自动推进
  • 指令中引用的支撑文件必须存在于指定路径——文件缺失会导致该步骤被静默跳过
Agent 只在指令明确引用时才加载支撑文件。
  • 确认指令中的文件路径与实际位置一致(例如 templates/chapter-outline.md,而不是 chapter-outline.md
  • 确认文件位于技能目录下正确的子目录中(templates/docs/scripts/
  • 如果 SKILL.md 中没有按文件名引用该文件,Agent 就没有理由加载它
项目技能始终优先于同名的全局技能。
  • 这是预期行为——如果项目技能处于激活状态,同名的全局技能会被完全忽略
  • 如果需要全局技能生效,请重命名或删除项目技能
  • 如果两者都需要存在,给它们起不同的名字并更新各自的描述

更多高级技巧

Workflows(工作流)

用工作流文件自动化可重复的多步骤流程

Rules(规则)

将你的写作偏好写入规则,让 Agent 始终用你的风格写作

Rule、Workflow、Skill 怎么选

三种 AI 指令方式各管各的事——了解它们的区别和适用场景