ai-crush-simulator（AI 粉碎模拟器）

AI Crush Simulator

What This Skill Does

AI Crush Simulator is a fun, safe, and encouraging skill for anyone navigating the classic mystery of figuring out a crush. It helps users:

1. Analyze a crush situation — read the signals, identify green/yellow/red flags,

and get a grounded vibe score.

1. Decode a text — interpret what a message might mean (with multiple possible

readings, never false certainty).

1. Generate reply options — get three thoughtful replies (Bold / Chill / Safe)

tailored to the user's goal.

1. Decide the best next move — get a clear, empowering recommendation with

practical tips.

Tone & Voice

• Playful and witty — this should feel fun, not clinical.
• Encouraging — celebrate the user's feelings without creating anxiety.
• Honest and grounded — never claim certainty about another person's feelings.
• Safe and respectful — never encourage pressure, manipulation, or harassment.

Every output must include a disclaimer framed as "based on what you shared..." to remind users these are observations, not facts about another person.

Modules

1. crushAnalysis — Situation Analysis

Input: CrushSituation

howTheyMet           string   // "class", "app", "mutual friend"
howLong              string   // "2 weeks", "6 months"
interactionFrequency string   // "daily texts", "occasional likes"
recentInteractions   string   // free-text description
yourFeelingConfidence 1–5     // how sure the user is about their feelings

Output: AnalysisResult

vibeScore       0–100    // heuristic composite score
connectionDepth          // surface | friendly | warm | potentially-romantic
flags           Flag[]   // green / yellow / red flags with reasons
signals         string[] // human-readable signal list
summary         string   // formatted multi-line summary
disclaimer      string

Logic:

• Scores against positive/negative/ambiguous signal dictionaries
• Detects named flags (e.g. "texted first" → green, "left on read" → red)
• Connection depth determined by score + relationship length
• Always neutral framing — no claim of certainty

2. textDecoder — Text Message Decoder

Input: TextInput

messageFromCrush  string   // the actual message
contextNote       string?  // optional extra context

Output: DecodedText

readings      TextReading[]  // up to 3 possible interpretations
overallVibe   string         // emoji + label e.g. "😄 Playful"
warmthScore   0–100
disclaimer    string

Logic:

• Scores warmth via enthusiasm markers, coldness markers, question count, length, emoji
• Builds 2–3 readings depending on warmth level and message characteristics
• Each reading has a confidence level (low / medium / high) and a vibe tag
• Multiple readings = honest reflection of ambiguity

3. replyGenerator — Reply Generator

Input: ReplyContext

decodedText  DecodedText
userGoal     keep-talking | show-interest | play-cool | ask-out
tonePref     funny | sincere | neutral  (optional)

Output: ReplyOptions

goal     UserGoal
replies  ReplyOption[]   // always exactly 3: bold, chill, safe
tip      string          // one coaching tip for this goal

Logic:

• Template-keyed by goal, delivering three tones: Bold / Chill / Safe
• Each reply includes a rationale so the user understands *why* it works
• No manipulative tactics (e.g. artificial scarcity, jealousy games, love bombing)
• Tips frame the user as confident and capable, not desperate or scheming

4. nextMove — Next Move Advisor

Input: MoveContext

vibeScore           number?    // from crushAnalysis
warmthScore         number?    // from textDecoder
userGoal            UserGoal?
howLong             string?
recentInteractions  string?

Output: NextMoveResult

action      keepChatting | askToHang | giveSpace | beMoreDirect | waitAndSee
headline    string   // emoji + short label
reasoning   string   // 2–3 sentence explanation
tips        NextMoveTip[]  // 3 practical, empowering suggestions
disclaimer  string

Decision Logic:

Heuristics Scoring Engine (heuristics.ts)

The scoring engine is shared across all modules and provides:

• scoreSignals(input) — scores a situation description 0–100 against 30+ signal patterns
• scoreTextWarmth(text) — scores a text message 0–100 for warmth/engagement
• detectFlags(input) — returns typed flags (green/yellow/red) from pattern matching

Positive signals (+8 each): texted first, remembered details, made plans, complimented, inside joke, quick replies, long replies, asked questions, followed up, invited, etc.

Negative signals (−10 each): left on read, one-word replies, cancelled plans, ignored, ghosted, never asks questions, avoids hanging, etc.

Ambiguous signals (−2 each): busy, might be shy, hard to read, late replies, emoji only, etc.

All scores are clamped to [0, 100].

Safety Constraints

The following constraints are enforced in every output:

1. No certainty claims — never say "they definitely like you" or "they don't like you". Always use "based on what you shared" framing.
2. No manipulation — never suggest jealousy games, ignoring messages on purpose, playing hard to get through deception, love bombing, or any pressure tactics.
3. No harassment — never encourage repeated contact after a clear rejection signal, following someone, or any behaviour that could constitute harassment.
4. No sexualized content — this skill is youth-friendly (13+). Zero sexual content in any output.
5. No stalking/surveillance — never suggest tracking someone's activity, checking their location, or monitoring their social media in unhealthy ways.
6. Empowerment framing — all advice is framed around the user's own confidence, self-respect, and authenticity. Never shame the user for their feelings.
7. Disclaimer required — every module output includes a disclaimer reminding the user that only the other person knows their true feelings.

Example Interactions

Example 1: Quick Text Decode

User: "My crush just texted me 'haha yeah' — what does that mean?"

Skill routes to: textDecoder

Output:

💬 Text Decoded
Overall vibe : 😐 Neutral
Warmth score : 22/100

Readings:
  1. [low confidence] neutral — Short reply — could be busy, tired, or
     not sure how to respond. One message isn't the full story.
  2. [medium confidence] friendly — Hard to gauge deeper intent from
     this alone — context from the broader conversation matters a lot.

ℹ️ Based on what you shared — texts can mean a lot of different things.
   These are possible interpretations, not facts about what they feel.

Example 2: Full Crush Check

User: "Can you do a full crush check? We met at a party 3 months ago..."

Skill routes to: Full flow (all 4 modules in sequence)

Example 3: Ask-Out Advice

User: "I want to ask my crush out, they just texted me about hanging out — give me options"

Skill routes to: replyGenerator with goal = ask-out

File Structure

ai-crush-simulator/
├── src/
│   ├── types.ts                    # All shared TypeScript interfaces
│   ├── modules/
│   │   ├── crushAnalysis.ts        # Module 1
│   │   ├── textDecoder.ts          # Module 2
│   │   ├── replyGenerator.ts       # Module 3
│   │   └── nextMove.ts             # Module 4
│   ├── scoring/
│   │   └── heuristics.ts           # Shared scoring engine
│   └── index.ts                    # Public API re-exports
├── cli/
│   └── main.ts                     # Interactive local CLI
├── tests/
│   └── scenarios.ts                # 3 sample test scenarios
├── SKILL.md                        # This file
├── README.md
├── package.json
└── tsconfig.json

Publishing Notes for ClawHub.ai

• All outputs are structured and screenshot-friendly — consistent formatting with clear section headers, emoji labels, and concise text.
• Modules can be called independently or composed into a full "crush check" flow.
• The skill has no external API dependencies — it runs entirely on heuristics, making it fast, private, and suitable for on-device or edge deployment.
• The heuristics engine can be extended by adding entries to the signal dictionaries in heuristics.ts without changing any module logic.
• Content is appropriate for ages 13 and up with G-rated output enforced in all templates.