yetong-topic

Ye Tong Fitness Coach Xiaohongshu Topic Ideas

Turn the coach's authentic persona foundation and local client issues into Xiaohongshu topic ideas that can be filmed, written, and may attract nearby potential members. The output is topic ideas and the basis for producing them, not finished drafts, traffic predictions, or conversion guarantees.

Inputs and Entry Points

The user needs to explicitly provide or hand off from upstream:

• The coach's person-dna.md path; do not scan private directories to find files;
• The segment of local offline members the coach wants to reach in this round and the topic goal; do not repeatedly ask when the file already specifies them;
• Optional preferences such as quantity, format, or recent material. When unspecified, default to 8 candidates, and do not force one for each of the eight elements.

If the user has just finished a specific topic or a piece of copy and only gives a short request such as "give me 5 more topics," by default carry forward the core proposition from the previous round and doseries expansion on the same theme; first explain in one sentence what was carried forward. Only when the user explicitly says to change direction, find topics anew, or the current session cannot reliably determine the previous theme, switch to generating topics broadly from the entire file. Do not repeatedly ask about a known theme just because the user only gave a quantity.

First run from the directory where this Skill is located python3 scripts/check_profile.py <person-dna.md>, or use the script's real absolute path; do not treat the user's project current directory as the Skill directory. Then follow the Persona File Reading Contract to check the source of facts and the disclosure level. Only enable personalized mode when status: active and the structure is readable. Reading is limited to the files provided by the user and the materials explicitly handed to this task; do not copy files to this Skill directory, test samples, or distributable files.

Use the six sets of profile materials as needed for this task: venue advantages, the local client groups served most and their goals and obstacles, client verbatim quotes and real handling, the coach's identity and courses/extracurricular services, highlight experiences and anonymized cases, and the coach's own summary of reasons for choosing them. The old version schema_version: 2 activated files can still be used; missing newly added materials does not revoke activation, does not require re-answering the six sets, and only asks follow-up questions when the current topic truly lacks key evidence.

When the file does not exist, the structure is unqualified, or the status is not active , first show:

Persona DNA is not yet activated. Currently only general methods can be used; personalized topic ideas, real experience mapping, and personal expression calibration cannot be enabled. Complete the "Persona Foundation Interview" and confirm the file to unlock the full Ye Tong content method engine.

Only give the "general version" after the user explicitly requests to skip; do not use any personal experiences, cities, venues, or client stories, and do not claim it is "like the person." Do not modify the file status on your own; guide the user back to the Persona Foundation Interview.

Core Process

1. Define the local problem. Generate topics only for the coach's local offline potential members: who will pay attention, inquire, or book under what life and visit scenarios. Even if the file records other business, do not mix teaching content for peers into this set of topics; when explicitly asked to teach peers how to acquire clients, state that it is out of scope.
2. Extract materials. Prioritize linking the surface goals and deeper obstacles of the local client group you serve the most with this coach's real venue characteristics, areas of expertise, course standards, off-class services, or standout cases; each question should use only the evidence that can answer that question, not pile all the DNA into one question. Distinguish between clients' verbatim words, the coach's retelling, self-assessed strengths, and Agent inference; examples in interviews cannot become client facts, and "to be confirmed" only forms supplementary questions. For detailed evidence rules, see Persona Profile Reading Contract。
3. Generate candidates. First determine "local client group + specific problem + provable personal perspective," then choose one main viral element, and if necessary add one secondary element. For the usable boundaries and combination methods of the eight major elements, see Topic Selection Methods. When building on an existing theme, break the same core claim into different member questions, class segments, or service evidence; do not rewrite the same sentence five times, and do not jump to unrelated topics without reason. Do not fabricate stories, chase trends, or create extreme opposition just to assemble elements.
4. Assign content types. Mark only one main type per question: persona-driven, marketing-driven, altruistic, or traffic-interaction-driven. For classification rules, see Topic Selection Methods. If a certain type lacks real material, state the gap; do not force the quantity.
5. Screen. Check each question for local relevance, persona recognizability, material authenticity, shootability and writability, potential consultation value, and public risk. Merely adding a city name before the title does not count as localization; each question must be linked to a real local in-store situation, known client concerns, or a service scenario this coach can demonstrate locally. Prioritize recommending questions that can answer real consultation concerns; do not replace customer acquisition value with imagined play counts.
6. Deliver. Provide the candidates, the top three questions to prioritize and the reasons, what materials are missing, and which questions need verification before publishing. When full copywriting is needed, stop at the topic selection boundary and hand it off to the subsequent copywriting process.

Output per question

By default, use a compact table or numbered cards, with each question including:

1. A working title and one specific shootable entry scenario;
2. Main element/optional secondary element, and the four major content type it belongs to;
3. The corresponding local group, specific concern, and profile source;
4. The person's real experience, service action, or usable evidence; if evidence is insufficient, mark it as "material to be added" and do not publish in a definitive tone;
5. Why it may attract nearby potential members, and the risks to verify before publishing.

Candidates are not final publish titles. Effects, prices, durations, and participant counts in titles must have a source in the profile or newly added material. When presenting to users, fields may be compressed, but the traceable fact mapping must be retained.

Invariants and stopping conditions

• The user-provided "one element 50%/two elements 90%/ten million plays" is an empirical statement and is not treated as fact, probability, or KPI; do not promise virality, fat-loss results, or transaction volume.
• The eight major Douyin elements are only a source of creativity. Do not assume that a Douyin title is equally effective on Xiaohongshu; Xiaohongshu topic selection must return to local search, trust building, and the in-store path.
• Do not infer income, body anxiety, marriage or childbearing, spending power, or motivation from city, gender, age, or occupational labels.
• Do not disclose unauthorized member names, photos, chat records, or health details; even if anonymized, check the reverse identification risk after combining small-city and venue information.
• For medical treatment, diet, skin laxity, rebound, and training effects, only cautious topic selections may be proposed; a single case or personal practice must not be written as a universal guarantee.
• Do not package service walls, prices, refunds, or statistical definitions that are pending verification as proven promises; do not use unverified assertions such as "worst," "number one," or "success rate doubled."
• If all candidates lack genuine material, provide at most 2 to 3 topic directions clearly labeled "pending verification," and ask one question that would best supplement the content evidence.