Nook Title

1 nook-title

Use this skill to turn article facts into high-click Chinese titles without inventing facts. The user does not need to name a source style; infer the appropriate AI-media title style from the task and facts.

This skill is a title system, not a slogan generator. It must preserve the factual boundary of the source, extract concrete hooks first, then generate and score candidates.

1.1 Core Rules

• Generate from facts only. Do not invent numbers, rankings, model names, companies, release status, benchmark results, dates, or user impact.
• Prefer concrete entities over abstract concepts: model, company, product, person, benchmark, number, cost, time, capability, affected group.
• For AI news and high-density media titles, target about 28-42 Chinese display characters when possible. Do not force length if the fact needs more or less room.
• Use punctuation as pacing, not decoration. ! and , are common in high-density AI-media titles, but only use them when there is a real event, contrast, or result.
• Keep the strongest factual hook in the first half of the title.
• Produce multiple structural variants, not ten small wording changes.
• Always include a fact-boundary note when the source is weak, speculative, sponsored, or not directly available.

1.2 Required Reference Loading

• For AI news, model/company release, benchmark, tool update, or high-density media titles, read references/high-density-ai-news-patterns.md.
• For safety, exaggeration control, and fact-boundary checks, read references/title-safety.md.
• When calibrating examples or validating style, read references/sample-titles.md.

1.3 Workflow

1. Extract title facts before writing:

• entities: model/company/person/product;
• numbers: price, time, benchmark, scale, version, count;
• action: released, leaked, beat, updated, cut cost, failed, got blocked;
• conflict or contrast: old vs new, A vs B, cost vs capability, promise vs risk;
• consequence: who is affected and why it matters.

1. Mark facts as confirmed, source-claimed, or unclear.
2. Choose 2-4 title frames from references/high-density-ai-news-patterns.md.
3. Generate 8-12 candidates. Include at least:

• 3 high-impact AI-media variants;
• 3 steadier WeChat variants;
• 2 search-friendly variants.

1. Score candidates with these dimensions:

• fact fidelity;
• entity clarity;
• numerical hook;
• tension/contrast;
• reader relevance;
• click strength without hallucination;
• readability in two seconds.

1. Recommend 2-3 candidates, not one. Explain the tradeoff briefly.
2. If the user asks for a publish title, provide publish_title, publish_summary, and optional publish_tags.

1.4 Output Format

For candidate generation:

事实钩子：
- 实体：
- 数字：
- 动作/冲突：
- 后果/人群：
- 边界：

候选标题：
1. ...

推荐：
- 第 X 个：理由
- 第 Y 个：理由

For title review:

标题诊断：
- 最大问题：
- 可保留：
- 风险：

改写：
1. ...

1.5 Script Use

Use scripts/analyze_titles.py when the user provides a list of real titles and asks for pattern extraction or statistics.

Use scripts/score_titles.py when the user provides candidate titles and wants a deterministic first-pass check. The script scores surface features only; final judgment must still check source facts and reader fit.

Examples:

python scripts/analyze_titles.py titles.txt
python scripts/score_titles.py candidates.txt

1.6 Coordination With Other Nook Skills

• If the task is a full WeChat article, nook-wechat-writer owns angle, article structure, and full draft. This skill owns title generation and scoring.
• If final wording feels AI-like or too polished, use ../nook-humanizer-zh-review/SKILL.md after title facts and structure are settled.
• Do not package or format the article for publication. Publishing assembly belongs to ../nook-wechat-packager/SKILL.md.

Source & license

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

• Author: captain-nook
• Source: captain-nook/nook_skills
• License: MIT

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