# Qiaomu Xinzhiyuan Title

> |

- **Type:** Skill
- **Install:** `agentstack add skill-joeseesun-qiaomu-xinzhiyuan-title-qiaomu-xinzhiyuan-title`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [joeseesun](https://agentstack.voostack.com/s/joeseesun)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [joeseesun](https://github.com/joeseesun)
- **Source:** https://github.com/joeseesun/qiaomu-xinzhiyuan-title

## Install

```sh
agentstack add skill-joeseesun-qiaomu-xinzhiyuan-title-qiaomu-xinzhiyuan-title
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Qiaomu Xinzhiyuan Title

把用户给出的文章、新闻素材或原始标题改写成“新智元版”中文标题：高密度实体、数字钩子、强转折、AI 圈语感，但不捏造事实。

## Trigger

Use this skill when the user asks to:

- 把文章、新闻、草稿或标题改成新智元风格。
- 给 AI/科技内容起“新智元版标题”或微信公众号爆款标题。
- 让标题更像新智元、更抓人、更有 AI 媒体冲击力。
- 在 Qiaomu workflow 中批量生成或筛选新智元味候选标题。

## Do Not Trigger

- 用户要学术论文标题、报告标题、乔木博客标题、36 氪/量子位/机器之心等其他媒体风格。
- 用户明确要求克制、官方、SEO、产品公告、招聘 JD 或法律合规文案。
- 用户只是总结、翻译、校对新智元文章，并明确“不改标题”。

## Required Reference

Before rewriting, read [Xinzhiyuan Title Style](references/xinzhiyuan-title-style.md). It contains the data-backed style rules extracted from the local 2688-article workbook.

## Workflow

1. Parse the input into a fact card:
   - core entity: model, company, lab, person, paper, product, event
   - action: 发布、开源、泄露、登顶、融资、跳槽、翻车、突破、被裁、下架
   - hard hooks: numbers, rankings, speed, cost, benchmark, time, money, headcount
   - authority: OpenAI, Anthropic, Google, Meta, DeepMind, Nature, Science, top conference, top university, known researcher
   - stakes: who benefits, who is threatened, what changes, why now
2. Choose one primary angle. Priority order:
   - major AI entity or famous person
   - hard number or benchmark
   - conflict, reversal, surprise, leak, ban, job movement
   - global/China/ranking/first-time framing
   - practical consequence for users, developers, researchers, companies
3. Generate 5-7 headline candidates:
   - one recommended standard 新智元 version
   - one number-heavy version when numbers exist
   - one conflict/reversal version when the material supports it
   - one authority/ranking version when the material supports it
   - one question-hook version when uncertainty or suspense is real
   - one restrained version for lower-risk publishing
4. Score candidates with the checklist in the reference. If many candidates are close, optionally run:

```bash
python3 scripts/score_title.py "候选标题1" "候选标题2"
```

5. Return the recommended title first, then alternatives and a short fact boundary.

## Hard Rules

- Do not invent facts. Do not add “刚刚”, “全球第一”, “首个”, “Nature”, exact money, user count, benchmark, ranking, or named people unless present in the source or explicitly provided by the user.
- Keep the default title between 24 and 42 Chinese characters. A longer title is acceptable only when it carries multiple true hooks.
- Use `！`, `，`, `：`, `？`, and `「」` deliberately. Do not stack punctuation until the title becomes noisy.
- Preserve real English model/company names such as GPT-5, Claude, Gemini, DeepSeek, OpenAI, Anthropic, Meta, xAI, Nvidia.
- If the source is dry or low-stakes, make it “有新智元味但不过度炸裂”; use the restrained candidate as default.
- Never transform sensitive medical, financial, legal, safety, or personal-risk claims into exaggerated certainty.

## Output Format

Default output:

```markdown
推荐标题：...

备选：
1. ...
2. ...
3. ...

风格依据：...
事实边界：...
```

When the user asks for only one title, output only `推荐标题` plus one brief `事实边界` note if any claim may need source support.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [joeseesun](https://github.com/joeseesun)
- **Source:** [joeseesun/qiaomu-xinzhiyuan-title](https://github.com/joeseesun/qiaomu-xinzhiyuan-title)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-joeseesun-qiaomu-xinzhiyuan-title-qiaomu-xinzhiyuan-title
- Seller: https://agentstack.voostack.com/s/joeseesun
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.
