Persona Pack Builder
Build persona products as prompt packs first. Treat this skill as a product-packaging workflow for personality/style bundles, not as a real-person cloning workflow.
Output goal
Produce a reusable folder that usually contains:
- - INLINECODE0
- INLINECODE1
- INLINECODE2
- INLINECODE3
- INLINECODE4
- INLINECODE5
- optional INLINECODE6
- optional INLINECODE7
- optional INLINECODE8
If the user only needs a lightweight deliverable, collapse this to a single persona.md plus examples.md.
Workflow
1. Classify the product
Identify which product is being built:
- - Prompt pack: reusable persona for end users across AI platforms
- Persona system: pack with multiple modes, JSON config, and examples
- Builder kit: reusable templates for generating many persona products
- OpenClaw skill candidate: only when the user explicitly wants an OpenClaw-installable workflow
Default to prompt pack unless the user clearly asks for a skill, installer, or agent workflow.
2. Extract the persona shape
Define these before writing files:
- - positioning: what the persona feels like
- relationship: friend / lover / coach / creator / etc.
- tone: direct / soft / playful / cool / mature / etc.
- boundaries: what it must not do
- formats: which platforms or prompt slots it must fit
- monetization tier: lite / standard / pro / custom
When source material references a real person, abstract it into public-trait language. Do not claim identity, ownership, or exact replication of a real person.
3. Create the core files
Write the files in this order:
- 1.
SYSTEM_PROMPT.md — concise, model-facing instructions - INLINECODE12 — expanded human-facing spec
- INLINECODE13 — short replies and multi-turn examples
- INLINECODE14 — structured representation for products/apps
- INLINECODE15 — buyer/operator instructions
- optional
modes/*.md, SALES_COPY.md, INLINECODE18
Prefer fewer stronger examples over large amounts of repetitive filler.
4. Keep the pack sellable
Optimize for:
- - clear emotional effect
- consistent tone
- low cringe / low oiliness
- platform portability
- easy customization
The buyer should be able to answer:
- - What feeling does this persona create?
- Where do I paste it?
- How do I tune it?
- What makes it different from random prompts online?
5. Package for the right audience
If the buyer is a normal AI user, deliver a prompt pack.
If the buyer is an advanced OpenClaw user, optionally also create:
- - a skill folder with this workflow
- templates under INLINECODE19
- references under INLINECODE20
Do not force persona content itself into a skill unless the user specifically wants agent-side automation.
Writing rules
- - Use short, concrete language
- Keep the system prompt tighter than the human docs
- Keep examples natural; avoid repetitive catchphrases
- Avoid obvious customer-support phrasing
- Avoid direct celebrity impersonation or “I am X” framing
- Prefer style abstractions such as “top-star energy”, “direct but warm”, or “protective but restrained”
File guidance
SYSTEM_PROMPT.md
Include:
- - role and tone
- interaction rules
- forbidden styles
- response priorities
Keep it compact enough to fit platforms with smaller system fields when possible.
PERSONA_CORE.md
Include:
- - positioning
- traits
- relationship dynamic
- language style
- emotional logic
- naming/calling style
- dos and don'ts
examples/
Include at least:
- - 15-30 short replies
- 5-10 multi-turn dialogues
Cover likely emotional states and product-selling moments.
config.json
Represent:
- - name/version
- traits
- speech style
- modes
- boundaries
- optional tuning knobs
README.md
Explain:
- - what the pack is
- who it is for
- included files
- how to use it
- what not to claim publicly
Decision rule: prompt pack vs skill
Choose prompt pack when the user wants to sell or use a persona directly.
Choose skill when the user wants an OpenClaw workflow that repeatedly generates persona packs, audits them, or standardizes packaging.
If both are useful, build the prompt pack first and the skill second.
Bundled resources
assets/templates/
Use the templates in assets/templates/ when the user wants a ready-made deliverable set. They provide a starter pack for:
- - INLINECODE22
- INLINECODE23
- INLINECODE24
- INLINECODE25
- INLINECODE26
- INLINECODE27
- INLINECODE28
Copy and customize them instead of rewriting the same scaffolding from scratch.
scripts/
Use scripts/generate_persona_pack.py to generate a starter persona-product folder from templates.
Example:
CODEBLOCK0
To override defaults, create a JSON file with replacement keys and pass:
CODEBLOCK1
Use the script when speed and packaging consistency matter more than bespoke wording.
References
- - Read
references/persona-product-blueprint.md for a compact packaging blueprint. - Read
references/safety-positioning.md when the request is inspired by a public figure or risks identity confusion. - Read
references/openclaw-skill-variant.md when converting the persona-product workflow into an installable OpenClaw skill.
角色包构建器
首先将角色产品构建为提示包。将此技能视为个性/风格组合的产品打包工作流程,而非真实人物克隆工作流程。
输出目标
生成一个可复用的文件夹,通常包含:
- - README.md
- SYSTEMPROMPT.md
- PERSONACORE.md
- config.json
- examples/shortreplies.md
- examples/conversations.md
- 可选 modes/*.md
- 可选 SALESCOPY.md
- 可选 FAQ.md
如果用户只需要轻量级交付物,则将其简化为单个 persona.md 加 examples.md。
工作流程
1. 对产品进行分类
识别正在构建的产品类型:
- - 提示包:跨AI平台面向最终用户的可复用角色
- 角色系统:包含多种模式、JSON配置和示例的包
- 构建工具包:用于生成多个角色产品的可复用模板
- OpenClaw技能候选:仅当用户明确要求OpenClaw可安装工作流程时
默认选择提示包,除非用户明确要求技能、安装程序或代理工作流程。
2. 提取角色形态
在编写文件前定义以下内容:
- - 定位:角色给人的感觉
- 关系:朋友/恋人/教练/创作者/等
- 语气:直接/柔和/俏皮/酷/成熟/等
- 边界:绝对不能做的事情
- 格式:必须适配的平台或提示槽位
- 变现层级:精简版/标准版/专业版/定制版
当源材料涉及真实人物时,将其抽象为公共特质语言。不得声称身份、所有权或对真实人物的精确复制。
3. 创建核心文件
按以下顺序编写文件:
- 1. SYSTEMPROMPT.md — 简洁的、面向模型的指令
- PERSONACORE.md — 扩展的面向人类的规范
- examples/.md — 简短回复和多轮对话示例
- config.json — 面向产品/应用的结构化表示
- README.md — 买家/操作者说明
- 可选 modes/.md、SALES_COPY.md、FAQ.md
优先选择数量较少但质量更高的示例,而非大量重复的填充内容。
4. 保持包的可销售性
优化方向:
- - 清晰的情感效果
- 一致的语气
- 低尴尬感/低油腻感
- 平台可移植性
- 易于定制
买家应能回答以下问题:
- - 这个角色营造什么感觉?
- 我该粘贴到哪里?
- 如何调整它?
- 它与网上随机提示有何不同?
5. 为正确受众打包
如果买家是普通AI用户,交付提示包。
如果买家是高级OpenClaw用户,可选择额外创建:
- - 包含此工作流程的技能文件夹
- assets/ 下的模板
- references/ 下的参考文件
除非用户特别要求代理端自动化,否则不要强制将角色内容本身放入技能中。
编写规则
- - 使用简短、具体的语言
- 保持系统提示比人类文档更紧凑
- 保持示例自然;避免重复的口号
- 避免明显的客服措辞
- 避免直接模仿名人或使用我是X的框架
- 偏好风格抽象,如顶级明星气场、直接但温暖或保护但克制
文件指南
SYSTEM_PROMPT.md
包含:
尽可能保持紧凑,以适应系统字段较小的平台。
PERSONA_CORE.md
包含:
- - 定位
- 特质
- 关系动态
- 语言风格
- 情感逻辑
- 称呼/叫法风格
- 该做与不该做
examples/
至少包含:
覆盖可能的情感状态和产品销售场景。
config.json
表示:
README.md
说明:
- - 包是什么
- 面向谁
- 包含的文件
- 如何使用
- 不得公开声称的内容
决策规则:提示包 vs 技能
当用户想要直接销售或使用角色时,选择提示包。
当用户想要一个能重复生成角色包、审核它们或标准化打包的OpenClaw工作流程时,选择技能。
如果两者都有用,先构建提示包,再构建技能。
捆绑资源
assets/templates/
当用户想要现成的交付物集合时,使用 assets/templates/ 中的模板。它们为以下内容提供入门包:
- - README.md
- SYSTEMPROMPT.md
- PERSONACORE.md
- config.json
- examples/shortreplies.md
- examples/conversations.md
- SALESCOPY.md
复制并定制它们,而非从头重写相同的框架。
scripts/
使用 scripts/generatepersonapack.py 从模板生成入门角色产品文件夹。
示例:
bash
python3 scripts/generatepersonapack.py --output /tmp/my-pack
要覆盖默认值,创建一个包含替换键的JSON文件并传入:
bash
python3 scripts/generatepersonapack.py --output /tmp/my-pack --values values.json
当速度和打包一致性比定制措辞更重要时,使用该脚本。
参考
- - 阅读 references/persona-product-blueprint.md 获取紧凑的打包蓝图。
- 当请求受公众人物启发或存在身份混淆风险时,阅读 references/safety-positioning.md。
- 当将角色产品工作流程转换为可安装的OpenClaw技能时,阅读 references/openclaw-skill-variant.md。