Xiaohongshu Title

Create Xiaohongshu Titles

Create a truthful reason to open the post. Build titles from the user's actual scene, tension, decision, method, evidence, and search language—not from fabricated stories or formula stacking.

Mode selection

• Direct mode: The user asks for a specific number of titles. Return that number and follow all length or tone constraints.
• Exploration mode: The user wants options or directions. Generate distinct strategic angles, then recommend one.
• Diagnosis mode: The user provides existing titles. Diagnose the precise weakness and rewrite it in the same direction and alternative directions.

Do not default to a large title dump when the user wants two or three usable options.

Input extraction

Identify:

1. target reader and immediate situation;
2. content job: search answer, execution method, case, comparison, opinion, proof, or conversion;
3. primary keyword when search discovery matters;
4. strongest real detail: action, object, number, constraint, mistake, trade-off, or outcome;
5. body promise and evidence available;
6. tone, format, title-count, and length constraints.

Do not invent price, sales, followers, revenue, time, identity, experience, reviews, product results, platform policy, or third-party proof.

Title angles

Use only angles supported by the content:

• audience + concrete problem;
• scene + unresolved tension;
• costly mistake + correction;
• counter-intuitive judgment + reason;
• number/checklist + execution object;
• comparison + decision rule;
• question phrased in the reader's language;
• experiment/case + method + boundary;
• short cover statement that creates recognition;
• search keyword + clear outcome or task.

Vary sentence structure. Do not generate synonyms of the same title.

Quality rules

• Let the reader recognize the applicable situation quickly.
• Keep one promise per title.
• Preserve the primary keyword naturally when search matters; clarity is more important than stuffing.
• Leave enough unresolved value to justify opening, but do not use fake suspense.
• Make the title consistent with the cover and body.
• Keep titles within the user's specified limit. When the user requests a Xiaohongshu-native short-title constraint, count characters and prefer 6–18 Chinese characters; do not present this as an official platform rule.
• Use first person only when the source contains a real first-person experience and the requested account voice allows it.
• Avoid empty superlatives, guaranteed results, fake urgency, exaggerated scarcity, and sensitive-category promises.

Style-risk filter

Rewrite titles that depend on formulaic filler such as “天花板、封神、闭眼入、被问爆、所有人都该、没人告诉你、建议收藏、狠狠拿捏” unless the user's verified voice and evidence clearly justify the wording. Treat this as an editorial filter, not a claim about platform policy.

Also reject repetitive contrast templates such as “不是 X 而是 Y” when the contrast adds no new information.

Output

Direct mode

Return the requested number of titles. When useful, group them by angle and mark the recommended title. Do not add an essay about the process.

Exploration mode

Return:

1. three to five title directions and when each fits;
2. distinct candidates under each direction;
3. one recommended title;
4. one sentence explaining the choice;
5. the primary keyword and any promise-risk note.

Diagnosis mode

Return:

1. the original title's specific problem;
2. three same-direction rewrites;
3. five alternative-direction rewrites;
4. one recommended title and reason;
5. any unsupported claim or body-delivery risk.

Final check

Before returning, verify that every title comes from the supplied facts, differs materially from the others, has a clear subject, matches the desired reader, and can be fulfilled by the body.

Source & license

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

• Author: ShenyuanNext
• Source: ShenyuanNext/wepr-growth-skills
• License: MIT

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