Social Signal and Tact

SkillCommunication

Read relationship signals across a conversation, infer what changed between turns, then turn awkward, overly formal, blunt, or AI-sounding messages to teachers, leaders, mentors, senior colleagues, partners, acquaintances, or potential professional contacts into short, natural, relationship-aware communication. Use for tactful praise, upward communication, progress updates, favors, reminders, apologies, disagreement, gift and courtesy messages, proxy greetings, holiday wishes, warm friend-like chat, trust-building, private invitations, role entrustment, third-party endorsement, and deciding when not to convert goodwill into a request.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the Social Signal and Tact skill

What this skill tells your AI

The instructions your AI receives, as published by fairy123456789/human-edge-agent-skills in skills/social-signal-and-tact/SKILL.md and read by ahel’s review.

Read the relationship before writing the sentence.

Core idea

Good “人情世故” is the ability to understand what the other person cares about, protect both sides' face, make appreciation concrete, and let the real request or sentiment arrive with the right emotional temperature.

For chat, read the interaction as a sequence of turns, not only as one finished paragraph. Sometimes the most natural message is shorter because part of the context should wait until the other person responds.

The skill should solve two tasks in order: first determine what the other person's words changed in the relationship; then decide how much of that change to acknowledge in the next message. Do not draft before the signal reading is clear.

Read signals before drafting

For any conversation with meaningful relationship stakes, silently reconstruct the interaction using these fields:

  1. Literal content — what was explicitly said or requested.
  2. Relational move — praise, trust, role assignment, face-saving, invitation, disclosure, endorsement, obligation, testing, distancing, or a concrete offer.
  3. Evidence strength — whether the signal is a single polite phrase, repeated across turns, public in front of third parties, costly in time or reputation, or backed by a concrete action.
  4. Role change — whether the user is being treated as a student, peer, trusted junior, collaborator, helper, family-adjacent guest, or potential professional contact.
  5. Response objective — receive the signal, protect status balance, deepen trust, keep optionality, make a practical next step, or deliberately leave the resource question untouched.
  6. Overreach risk — what the user would be assuming if they responded too strongly.

Use cumulative evidence. Repeated specific praise is stronger than one compliment. Public advocacy in front of relevant third parties is stronger than private praise. An invitation into private or family space usually signals more relational warmth than a generic “有空联系”. A concrete introduction, meeting arrangement, document request, or work assignment is stronger evidence of resource intent than encouragement alone.

Do not collapse different signal types into one conclusion. Personal warmth can coexist with professional uncertainty. Praise can be sincere without implying a job, internship, introduction, or future commitment.

A useful internal scale is:

  • L0 — Courtesy: greetings, thanks, routine politeness.
  • L1 — Personal recognition: specific praise, remembered detail, voluntary encouragement.
  • L2 — Trust / role entrustment: asking the user to help, advise, accompany, represent, or look after someone; repeated or public endorsement also belongs here.
  • L3 — Private-circle opening: home-town invitation, family context, personal disclosure, reciprocal hospitality, or inclusion in a more private social setting.
  • L4 — Resource activation: a concrete introduction, invitation to a work setting, request for a résumé or proposal, named follow-up owner, scheduled meeting, actual task, or explicit offer.

Treat the scale as evidence discipline, not a rigid social ranking. The next message usually matches the observed level or stays slightly below it. Move higher only when the other person has opened that layer first or when a concrete reciprocal gesture is natural.

Relationship pacing rules

When a senior person gives the user a higher-status role over a peer, protect both relationships. For example, if a parent says “please help him become more disciplined,” do not accept a supervisory posture too literally. Reframe it as 互相督促 / 一起进步 / 能帮上的尽力 unless the user truly has formal responsibility. This receives the elder's trust while preserving the peer's dignity.

When praise becomes identity-level—such as maturity, judgment, execution, or future potential—receive it with modesty and one grounded sentence. Do not argue with the praise, amplify it into self-promotion, or immediately attach a request.

When the other person opens a private or family layer, deepen only with true specifics. A reciprocal invitation can be more natural when it names a real place, food, activity, or shared interest. Specificity turns a ceremonial “欢迎来玩” into a believable future social scene. Use this only when the warmth has already been opened by the other person.

Do not cash out goodwill too early. A compliment, public endorsement, or invitation can increase trust without yet becoming a professional resource. Preserve the relationship until there is an L4 trigger or a separate, naturally justified reason to ask. This is especially important when the same network may later contain internships, introductions, collaboration, or hiring opportunities.

When the other person's message reveals two identity anchors, respond to the one that matters most to them. A senior may care about professional expertise and family responsibility at the same time. A short, specific acknowledgement of both can show that the user actually listened. Avoid inventing emotional depth that the conversation did not support.

Workflow

  1. Identify the relationship: teacher, leader, mentor, senior colleague, partner, acquaintance, potential professional contact, or real friend.
  2. Infer the immediate social goal: report progress, ask for help, praise, thank, remind, repair awkwardness, disagree, give a gift, pass on greetings, send a holiday wish, make a request, receive trust, or respond to a relationship upgrade.
  3. Reconstruct the signal sequence using literal content → relational move → evidence strength → role change → response objective → overreach risk. If the user asks what the other person “really means,” answer this before drafting.
  4. Check the speaking perspective, especially with teachers, mentors, senior people, or benefactors. Build from what the user saw, remembered, is doing now, or genuinely appreciates. Do not evaluate the other person's life, summarize their personality, infer their habits, or arrange their future. Even when the user supplies an observable detail, mention it neutrally from the user's perspective instead of converting it into a judgment about the recipient.
  5. Separate observable facts from interpretation. Praise should attach to something real: a decision, useful feedback, a result, a way of thinking, a resource, or a concrete effect on the user. In gratitude messages, one small true memory—an action, a sentence, or one concrete act of help—is usually enough.
  6. Estimate relational distance, signal level, evidence confidence, and channel. A WeChat message to a familiar teacher should not read like an official email. Do not infer L4 resource activation from L1–L3 warmth alone.
  7. Identify the other person's likely priority: progress, respect, control, efficiency, recognition, low-risk execution, being understood, family concern, trust, or receiving a gesture without social pressure.
  8. Decide whether the user should receive, mirror, slightly under-match, or defer the signal. For upward relationships, under-matching by a small amount is often safer than escalating faster than the other person.
  9. Choose one pattern from references/message-patterns.md. Use multiple techniques only when the message still feels natural. For teacher or mentor holiday gratitude, use the dedicated pattern there.
  10. Decide whether the content belongs in one send or several chat turns. Split WeChat messages by social function, not merely for visual formatting. If the situation contains both immediate logistics and relational explanation, lead with the immediate fact, then add the relationship context in a second message or after a natural reply. Avoid front-loading delivery details, apology, family context, product explanation, and well-wishes into one dense block.
  11. For gifts or courtesy visits, describe enough to make the gesture understandable, then stop. State the source, timing, local or seasonal quality, or caring intention when relevant. Do not sell the gift through price, brand prestige, rankings, exaggerated scarcity, or medical claims.
  12. Match the weight of the apology to the actual situation. For a minor absence or missed visit, phrases such as 有些过意不去 and 让我代他向您问好 are often proportionate. Reserve 请您见谅 or stronger apology language for situations that genuinely require repair.
  13. Make senior-facing closing language invitational rather than directive. Prefer 您回办公室后可以尝尝呀,希望您吃着喜欢 over commands such as 记得拿一下 when the context is a gift.
  14. Never frame a senior, mentor, or benefactor as someone who only came to mind because of the occasion. Avoid phrases such as 很久没联系了, 正好借教师节向您问个好, 顺便跟您问候一下, or 刚好想到您. Enter from the relationship itself: a direct blessing, gratitude, one real memory, or a current update.
  15. Keep each send short by default. In chat, one compact paragraph often works better than a structured memo, and two or three functionally distinct messages can work better than one overloaded paragraph.
  16. Remove AI markers: generic praise, repeated thanks, over-explaining background, emotional overstatement, bureaucratic transitions, three-stage mini-essays, and ornamental parallel endings such as 把 A,把 B or 做好 A,走稳 B. Stop once one natural meaning is complete; do not add a second action merely for rhythm.
  17. Make the other person easy to respond to. Requests should be clear and low-friction, with room for timing or refusal when appropriate.
  18. If professional optionality exists, check whether the current turn truly contains an L4 trigger. If it does not, do not force a résumé, internship, job, collaboration, or introduction request into a relationship-building exchange.
  19. Run the final check below.

Final check

  • What changed in the relationship during the last turn?
  • Is that change supported by words only, repeated behavior, public endorsement, costly action, or a concrete next step?
  • Am I treating warmth as a resource commitment without evidence?
  • Is the user being assigned a role over a peer that should be softened into mutual support?
  • Would a real person send this, and would they send all of it at once?
  • Does any sentence make the senior person feel evaluated by the user?
  • Does it contain an inference the user did not explicitly provide?
  • Could it make the recipient feel remembered only because a holiday arrived? In holiday greetings, remove phrases such as 很久没联系了, 正好借教师节问个好, or 顺便问候一下.
  • Did the draft add an unnecessary sentence for polish, symmetry, or a sense of completeness?
  • Should this WeChat message be split into multiple bubbles with distinct social functions?
  • If the relationship just became warmer, does the response receive that warmth without immediately turning it into a favor or professional ask?

Tone controls

The user may ask for one of these:

  • 克制礼貌
  • 熟一点但不冒犯
  • 自然拍一下马屁
  • 汇报中顺手归功
  • 请教式
  • 轻松朋友感
  • 有分寸地反对
  • 送礼/代问候
  • 节日感谢
  • 潜在人脉关系维护
  • 关系信号拆解
  • 先养关系,暂不提资源

Friend-like naturalness is allowed when the wording fits the real relationship. Fabricated intimacy is not.

Integrity guardrail

Do not invent shared memories, private closeness, endorsements, affection, resource commitments, or facts. Do not coach deception that materially misrepresents the user's relationship or situation. For supplements, health products, or wellness gifts, avoid unsupported efficacy claims and status claims; keep the message focused on care and the actual gift.

Signals

GitHub stars
90
Last commit
Sep 2026
Advanced
Item type
skill
Key
social-signal-and-tact
Source
github.com/fairy123456789/human-edge-agent-skills