Skip to content

Latest commit

 

History

History
152 lines (119 loc) · 3.82 KB

File metadata and controls

152 lines (119 loc) · 3.82 KB

Communication Patterns Analysis - Data Index

Analysis of 1,352 messages from Jess's Claude Code sessions across 4 projects.


Raw Data Files

/tmp/jess_messages_raw.json (3.8 MB)

All 1,352 human messages extracted from conversation logs.

Structure:

{
  "project": "personal-log",
  "session": "df3f5123",
  "timestamp": "2025-12-04T02:07:11.977Z",
  "human": "Hey, something's gone wrong with my mobile app...",
  "assistant_before": null,
  "is_first": true
}

Fields:

  • project: Which project folder (personal-log, blog, pl-context, roguelike)
  • session: Session ID (first 8 chars of conversation filename)
  • timestamp: When the message was sent
  • human: The actual message content
  • assistant_before: Claude's message immediately before (for context)
  • is_first: Whether this was the first message in the session

/tmp/jess_messages_analyzed.json (1.8 MB)

Messages with computed characteristics.

Additional fields:

  • word_count: Number of words
  • char_count: Number of characters
  • starts_lowercase: Boolean
  • has_question: Contains "?"
  • has_ellipsis: Contains "..."
  • has_code: Contains backticks
  • references_previous: References earlier context

/tmp/jess_conversation_pairs.json (3.1 MB)

776 message pairs for cause-effect analysis.

Structure:

{
  "human": "yup that worked",
  "assistant_response": "Great! Now let's...",
  "human_category": "feedback",
  "response_type": "continuation"
}

/tmp/jess_communication_patterns.json (10 KB)

Structured analysis of identified patterns.

Contains:

  • meta: Total messages, projects, date range
  • communication_patterns: 8 patterns with frequencies, signal words, examples
  • style_characteristics: Lowercase starts, question frequency, ellipsis usage
  • notable_openers: Most common message starters

Teaching Documents

/tmp/agentic_ai_workflow_guide.md

Main practical guide: "Getting Effective Results with Agentic AI"

  • Step 1: Know what you're trying to accomplish
  • Step 2: Communicate to stay aligned (detail, validation, feedback)
  • Step 3: Recognize when you're stuck
  • Step 4: For complex work, use structure

/tmp/claude_communication_guide.md

Comprehensive guide with all 8 patterns, session arc, anti-patterns.

/tmp/claude_quick_reference.md

One-page cheat sheet.

/tmp/meta_blog_post.md

Blog post telling the story of this analysis session.


Projects Analyzed

Project Description
personal-log Mobile app development (React Native, Expo, Firebase)
blog Jekyll blog with syntax highlighting
pl-context Related tooling/infrastructure
roguelike Game project

Date range: 2025-12-04 to 2026-01-17


Key Statistics

  • Total messages: 1,352
  • Feedback loop messages: ~30%
  • Quick questions: ~16%
  • Thinking aloud: ~9%
  • Affirmation + pivot: ~9%
  • Course corrections: ~8%
  • Starts lowercase: 52%
  • Contains questions: 51%
  • Uses ellipsis: 29%
  • References previous context: 80%
  • Bare affirmations ("yes", "ok"): 0.1%

How to Query the Data

# Count messages by project
cat /tmp/jess_messages_raw.json | python3 -c "
import json, sys
from collections import Counter
data = json.load(sys.stdin)
print(Counter(m['project'] for m in data))
"

# Find messages containing a pattern
cat /tmp/jess_messages_raw.json | python3 -c "
import json, sys
data = json.load(sys.stdin)
for m in data:
    if 'your pattern' in m.get('human', '').lower():
        print(m['human'][:200])
        print('---')
"

# Analyze message lengths
cat /tmp/jess_messages_analyzed.json | python3 -c "
import json, sys
data = json.load(sys.stdin)
lengths = [m['word_count'] for m in data]
print(f'Mean: {sum(lengths)/len(lengths):.1f} words')
print(f'Max: {max(lengths)} words')
"