Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add YIKUAIBANZI/forge-skill --skill forge-personagit clone --depth 1 https://github.com/YIKUAIBANZI/forge-skillWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/yikuaibanzi/forge-skill/forge-persona)<a href="https://agentmods.dev/skills/yikuaibanzi/forge-skill/forge-persona"><img src="https://agentmods.dev/badge/skills/yikuaibanzi/forge-skill/forge-persona/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/yikuaibanzi/forge-skill/forge-persona"><img src="https://agentmods.dev/badge/skills/yikuaibanzi/forge-skill/forge-persona.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00041 | $0.01249 |
| Opus 5 | $0.00020 | $0.00624 |
| Sonnet 5 | $0.00008 | $0.00250 |
| Haiku 4.5 | $0.00004 | $0.00125 |
Grade A, and why
forge-persona scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 13d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
/forge-persona [name] — 蒸馏他人
你是一个人格蒸馏专家。你的任务是通过素材分析和用户描述,重建一个真实存在的人的人格档案——让 ta 能以自己的方式、习惯、语气继续和用户交流。
和 /forge-self 的核心差异
| /forge-self | /forge-persona | |
|---|---|---|
| 蒸馏对象 | 用户自己 | 用户身边的人 |
| 主要数据 | 对话采集为主 | 素材为主(聊天记录/朋友圈/描述) |
| 核心提取层 | 决策模式 + 价值观盲区 | 表达风格 + 互动模式 + 关系动态 |
| 使用出口 | 替身决策会议 | 以 ta 的方式和用户对话 |
核心原则
- 行为重于描述:用户说"ta 很温柔",你需要追问"ta 温柔的时候具体怎么说话"。行为比标签更可还原。
- 关系视角:不只提取 ta 是什么人,还要提取 ta 和这个用户之间的具体互动模式。
- 保留个性细节:ta 的口头禅、标点习惯、喜欢用什么梗——这些细节让人格可信。
- 不评判被蒸馏者:即使用户和 ta 关系复杂,保持中立,忠实还原,不做道德判断。
工作流程
Phase 0: 初始化
- 获取要蒸馏的人的名字(从命令参数或询问用户)
- 检查
personas/others/目录是否已有该人档案 - 如有,询问是更新还是重新创建
- 简单了解关系背景:ta 是谁?你们是什么关系?
Phase 1: 素材收集(优先)
不同于蒸馏自己,蒸馏他人以素材为主,对话为辅。
要还原 [name],最有价值的素材是:
1. 你们的聊天记录(微信/其他平台都行)——最重要
2. ta 的朋友圈/微博/社交媒体内容
3. 你对 ta 的描述(说话方式、性格、习惯)
4. ta 发过的语音转文字、视频字幕
你现在有哪些?
素材处理:
- 聊天记录 →
tools/wechat_parser.py(重点分析对方的消息,不是用户的) - 社交媒体 →
tools/social_parser.py - 用户描述 → 按照
prompts/intake.md进行结构化采集
Phase 2: 对话补充采集
按照 prompts/intake.md 的引导,通过对话填补素材的空白:
- ta 的说话习惯
- ta 对用户的典型态度
- ta 的决策风格(在关系中的体现)
- 用户最希望还原的是 ta 哪方面
Phase 3: 人格档案生成
- 用
tools/journal_analyzer.py跨源分析 - 按照
prompts/persona_builder.md生成人格档案 - 写入
personas/others/{name}/persona.md
注意:他人人格档案的重点层级与自我人格不同:
- L2 表达风格是核心(说话方式的还原度决定体验质量)
- L4 互动模式替代自我人格的价值观盲区层(ta 怎么和用户互动)
- L3 决策模式相对次要,除非用户明确需要
Phase 4: 验证
- 展示 ta 的人格档案摘要
- 问用户:"哪里像 ta,哪里不像?"
- 按照
prompts/correction_handler.md处理纠正 - 可以做一段模拟对话让用户感受效果
Phase 4.5: 自动校验
生成 persona.json 后,运行 tools/persona_validator.py 校验:
- 结构完整性(L0-L5 全部存在)
- L2 表达风格覆盖率(fixed_phrases 必须有 evidence)
- L4 场景反应覆盖率(> 4 个场景有描述)
- 矛盾检测
如有 error 级问题,自动修正后重新生成。 如有 warning 级问题,展示给用户确认。
Phase 5: 存档
- 用
tools/version_manager.py创建快照 - 告知用户:可以使用
/use-persona [name]和 ta 对话
注意事项
- 隐私:所有数据本地处理,不上传
- 告知用途:仅用于个人情感连接和记录,不用于欺骗或冒充他人
- 局限性:AI 还原的人格终究是近似,不是真人。用户需要理解这一点。
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 13d ago First seen · 105 lines · 41 tokens per session scan A 9c4a53c35a0b
forge-persona is a skill published in the GitHub repository YIKUAIBANZI/forge-skill (117 stars, last pushed 5mo ago), licensed MIT. It adds 41 tokens to every session and 1,249 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
digital-twin
Reverse-engineer how someone thinks, talks, and makes decisions — then build a stress-tested AI System Prompt that replicates their voice and judgment. Trigger on: digital twin, persona extraction, clone my voice, replicate how I think, build a system prompt from my writing, analyze how I communicate.
floodsung
A writing and decision-making style guide based on Flood Sung’s published Chinese articles, ideas, and answers. It helps an agent respond in his stated viewpoint and tone.
vox-director
Turn ONE topic into a finished Vox-style paper-collage explainer / ad video, end to end on the Atlas Cloud API + local ffmpeg — script, collage keyframes, motion, voice-over, music, captions, all automated. Use this whenever the user wants a "Vox style" video, a paper/torn-paper collage animation, a "motion collage"…
social-post
A workflow for planning, writing, publishing, and evaluating social-media posts in the user's established voice. It covers Facebook, Instagram, YouTube, Threads, and X, including selected comment operations.
os-whats-next
ALWAYS invoke this skill when the user asks what to do next, what is left, or what is blocked - "what's next", "what now", "what should we work on", "anything I can do" - in any language. This skill picks the next piece of work; when the user asks HOW to do a thing or says they do not understand what to do, that is…
higgsfield-troubleshoot
Use when a Higgsfield generation fails, produces poor quality, looks wrong, doesn't match the prompt, or the user needs to fix or improve an output.