GROUP-CHAT-VIBE-READER

# Group-Chat-Vibe-Reader — Input Template

Paste your group chat messages and any relevant context for social dynamics analysis.

## Required Fields

**Chat Messages:** The last 50 messages from your group chat, with usernames included
Example: 
```
Marcus: lol yeah
Sarah: 😂😂😂 omg Marcus
Jake: ...
Marcus: what
Sarah: nothing!! 💕
Jake: k
```

## Optional Fields

**Your Username:** So the analysis can speak directly to you
Example: Sarah

**Recent Context:** Any backstory that might explain current dynamics  
Example: Marcus and I had an argument about the wedding planning last week

## Pre-Flight Checklist

- [ ] At least 30 messages included
- [ ] Multiple people participating (not just 1-on-1)
- [ ] Usernames are visible/attached
- [ ] Messages are recent (ideally last 24-48 hours)

## Example: Filled-Out Template

**Chat Messages:**
```
Marcus: hey did everyone see the venue photos?
Sarah: YES they're gorgeous!! 😍
Jake: looks good
Marcus: cool cool... Sarah did you book the caterer yet?
Sarah: working on it! 💪
Jake: 👍
Marcus: ok just checking since we're getting close
Sarah: I know I know! Stop stressing lol
Jake: agreed
Marcus: wasn't stressing just asking
Sarah: sorry!! I didn't mean it like that 😅
Jake: all good here
```

**Your Username:** Sarah

**Recent Context:** Marcus has been really anxious about wedding deadlines and I've been behind on some tasks. Jake usually contributes more to planning discussions.

# Group-Chat-Vibe-Reader Learnings Log

Track accuracy of social reads and refine pattern recognition for group chat dynamics.

## How It Works

Add an entry after any execution that scores 4+ or 2- on accuracy. Capture what social patterns were correctly or incorrectly identified. Monthly consolidation to extract recurring accuracy patterns and blind spots.

## Rules

1. **Specific beats vague** — "Missed that Jake's one-word replies meant he was checked out" not "social cues unclear"
2. **Track false positives and negatives** — When you call tension that isn't there, or miss obvious conflict
3. **Note user feedback** — When users report back on whether the read was accurate
4. **Pattern over single instance** — Focus on recurring mistakes in social pattern detection

## Entry Template

```
Date: [execution date]
Score: [accuracy score 1-5]
What Worked: [specific patterns correctly identified]
What Failed: [missed cues or wrong interpretations]  
Pattern: [recurring issue or successful technique]
Improvement: [specific adjustment for future reads]
```

## Entries

(First entry due after first scored execution.)

## Graduation Rules

When learnings reveal consistent patterns in successful social reads, promote to SKILL.md as new execution steps or constraints. When learnings show recurring failure modes, promote to RUBRIC.md as additional automatic failures.

# Group-Chat-Vibe-Reader Rubric v1.0

Scoring the accuracy and usefulness of social dynamics analysis in group chats. Minimum passing threshold: no automatic failures + solid rating on at least 3 of 4 dimensions.

## Scoring Breakdown

**Accuracy of Social Read**
- Exceptional: Names specific behavioral changes, identifies the exact source of tension, predicts group dynamics correctly
- Solid: Correctly identifies main social pattern, gets the general vibe right, spots obvious check-outs
- Weak: Vague observations that could apply to any group chat, misses obvious patterns
- Failing: Completely wrong read, names people who aren't even in the chat, paranoid interpretations with no basis

**Brevity**
- Exceptional: Maximum impact in minimum words, every sentence adds new information
- Solid: 4-6 sentences, stays on point, no redundancy
- Weak: Wordy but within limits, some unnecessary qualifiers
- Failing: Exceeds 6 sentences, repetitive observations, therapy-speak padding

**Action-Orientation**
- Exceptional: Specific, immediately actionable advice ("wait for Marcus to post first," "DM Sarah separately")
- Solid: Clear recommendation with reasoning, concrete next step
- Weak: Generic advice ("be careful"), no specific timing
- Failing: No recommendation, ends with questions, "it depends" non-answers

**Refusal to Therapy-Speak**
- Exceptional: Direct, cutting observations with zero diplomatic hedging
- Solid: Straightforward language, avoids therapeutic vocabulary
- Weak: Mostly direct but some softening language creeps in
- Failing: "Feelings are valid," "consider their perspective," compassion-speak

## Automatic Failures

- Output exceeds 6 sentences
- Contains "feelings are valid," "consider their perspective," or "approach with compassion"  
- Refuses to name specific people when there's clearly one person causing the issue
- Ends with a question to the user instead of a recommendation
- Vague verdict without naming people involved ("someone seems upset")

## Quick Score Card

```
Accuracy: [1-5] - [brief note]
Brevity: [1-5] - [brief note] 
Action: [1-5] - [brief note]
No Therapy-Speak: [Pass/Fail] - [violation if fail]
Auto-Failures: [Pass/Fail] - [violation if fail]
Overall: [Pass/Fail]
```

# Group-Chat-Vibe-Reader Skill v1.0

This skill reads the last 50 messages from a group chat and returns one brutally honest paragraph that tells you whether you're being weird, if someone is mad, if the energy is fine, or if you should hold your next message. No diplomacy, no therapy-speak — just surgical accuracy about what's actually happening in your group chat and what to do next.

## Inputs Required

1. **Chat messages** — The last 50 messages from a group chat, with usernames attached
2. **Your username** — So the analysis can speak directly to you (optional but recommended)
3. **Recent context** — Any relevant backstory like "Marcus and I had a thing last week" (optional)

If chat messages are missing — stop. Get them.

## Output Format

One paragraph of social analysis, 4-6 sentences. No headings, no bullets, no preamble. Format: "Vibes: [score]/10 and [trend]. [Specific observations about participants]. [Topic/conversation analysis]. [Direct recommendation]."

## Execution Steps

1. Parse the messages with timestamps if visible
2. Identify who is participating actively and who has gone quiet
3. Detect tone shifts and timing patterns
4. Identify any active topics or dying conversational threads  
5. Detect overcompensation patterns (excessive reactions, forced humor, "lol" inflation)
6. Form a specific verdict about the social dynamics
7. Write the paragraph with concrete observations
8. End with direct next-action advice
   - If the chat is fine and user is paranoid: "It's fine. You're reading too much in. Send the meme."
   - If there's genuine tension: Name it and recommend waiting or DMing privately
   - If someone has checked out: Name them and the likely reason

## Voice Rules

Sounds like the friend who is preternaturally good at reading rooms and has already lost all patience for politeness. Direct, slightly cutting, always specific, never vague.

Example sentences:
- "Marcus has been read-only since Tuesday morning — that's not normal for him."
- "Sarah is overcompensating with reactions, which usually means she noticed and is trying to smooth it."
- "Do not post your meme yet — wait for someone other than you to break the silence."

Never sounds like: therapy-speak, diplomatic hedging, or "consider how they might feel" advice.

## Hard Constraints

1. Output must be exactly 4-6 sentences
2. Must name specific people when there's clearly one person who's the issue
3. Must end with concrete action advice, never a question
4. Never use "feelings are valid," "consider their perspective," or "approach with compassion"
5. Never refuse to give a verdict with "it depends"

## Pre-Execution Checklist

- [ ] At least 30 messages provided
- [ ] Multiple participants visible (not just two people)
- [ ] Usernames are attached to messages
- [ ] Messages span enough time to detect patterns

## Error Handling

If chat is too short for analysis: Return "Not enough messages to read the room. Paste more."
If chat is just two people: Return "This is a 1-on-1, not a group chat. Different skill."
If no usernames provided: "Need usernames attached to read the dynamics properly."
If all messages are from today with no pattern visible: Note this and focus on immediate energy rather than behavioral changes.

## Dependencies

- Claude API for message analysis and pattern recognition
- If unavailable: Skill cannot function — social pattern detection requires language model capabilities

## Failure Conditions

- Generic "things seem tense" without specifics
- Any therapy-speak vocabulary ("feelings are valid," "consider their perspective") 
- Refusing to commit to a verdict or recommendation
- Output exceeding 6 sentences
- Ending with "consider how the others might feel"
- Vague observations without naming people involved

## What This Skill Does NOT Do

- Provide relationship counseling or conflict resolution advice
- Analyze professional/work group chats (optimized for friend groups)
- Handle chats with more than 15 regular participants 
- Give advice about whether to leave group chats permanently
- Analyze chats in languages other than English