Claude Code Plugins

Community-maintained marketplace

Feedback

test-voice-report

@cncorp/arsenal
8
0

Test report generation with TEXT transcripts (NOT real calls). For real Twilio calls, use twilio-test-call instead.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name test-voice-report
description Test report generation with TEXT transcripts (NOT real calls). For real Twilio calls, use twilio-test-call instead.

test-voice-report Skill

STOP! Are you trying to place a REAL Twilio call?

This skill only simulates calls from TEXT files. It does NOT place real Twilio calls.

For real calls, use /twilio-test-call instead.

Want to... Use this skill
Test report generation with text test-voice-report (this one)
Place a real Twilio call twilio-test-call

🎯 Purpose

Test the voice call summary report system end-to-end with sample conversation transcripts.

Use this skill when:

  • Testing changes to the voice_call_summary_report Langfuse prompt
  • Validating report quality with different conversation patterns
  • Debugging report generation issues
  • Creating new test conversations

What This Skill Does

This skill provides a single command to:

  1. Create a simulated VOICE conversation from a transcript file
  2. Generate the Wren-style call report using the Langfuse prompt
  3. Send the report to the couple's GROUP conversation
  4. Show a preview of the generated report

Before this skill: Required running 2 separate scripts With this skill: One command, complete workflow

Quick Start

Test with existing transcript

docker compose exec api python src/scripts/test_voice_call_report.py \
  --transcript .claude/skills/test-voice-report/transcripts/yard_chores.txt \
  --duration 480

Create and test new transcript

# 1. Create new transcript file in arsenal (source of truth)
vim arsenal/dot-claude/skills/test-voice-report/transcripts/my_conversation.txt

# 2. Sync to .claude
./arsenal/install.sh

# 3. Test it
docker compose exec api python src/scripts/test_voice_call_report.py \
  --transcript .claude/skills/test-voice-report/transcripts/my_conversation.txt \
  --duration 300

Available Transcripts

Current test transcripts:

  • transcripts/yard_chores.txt - Pursuer discussing household chores (8 min call)

To add more:

  1. Create new .txt file in arsenal/dot-claude/skills/test-voice-report/transcripts/
  2. Add raw conversation text (no special formatting needed)
  3. Run ./arsenal/install.sh to sync to .claude/
  4. Test with the command above

Script Parameters

docker compose exec api python src/scripts/test_voice_call_report.py \
  --transcript PATH           # Required: Path to transcript file
  --duration SECONDS          # Optional: Call duration (default: 300)
  --group-conversation-id ID  # Optional: GROUP conversation ID (default: 10)
  --simple-parsing            # Optional: Use simple alternating instead of smart parsing

Parameters:

  • --transcript: Path to transcript text file
  • --duration: Call duration in seconds (affects call_duration_minutes in report)
  • --group-conversation-id: Which couple's GROUP conversation to use (default: 10 = camily & daniel)
  • --simple-parsing: Use simple sentence-by-sentence alternating instead of smart grouping

Transcript Format

Transcripts are very simple - just raw conversation text:

"I don't want to fight, but we need to talk about the yard."
"Is this about the dogs?"
"Yes. There's poop everywhere."
"I picked some up yesterday."

No special formatting needed:

  • ❌ No speaker labels
  • ❌ No timestamps
  • ❌ No metadata
  • ✅ Just the conversation text

Parsing modes:

Smart parsing (default):

  • Groups related sentences together
  • Switches speaker at questions
  • Detects response patterns ("Yeah", "Okay", "So")
  • More natural conversation flow

Simple parsing (--simple-parsing flag):

  • Alternates speaker every sentence
  • More predictable but less natural

What The Report Tests

The generated report should demonstrate:

✅ Attachment Pattern Identification

  • Pursuer or Withdrawer classification
  • Based on conversation patterns

✅ Ask Better / Listen Better Framework

  • Role-specific guidance (NOT generic Gottman)
  • Pursuers get "Ask Better" practices
  • Withdrawers get "Listen Better" practices

✅ Neuroscience Explanations

  • Arousal hijacking (>100 bpm = lose IQ points)
  • Connection under threat
  • Protest behavior framing

✅ Wren's Voice

  • Normalizing, warm, research-backed
  • "That intensity isn't 'too much'"
  • Reframes conflict as attachment system, not personal failure

✅ Specific Examples

  • Quotes from the actual call
  • Identifies Ask Better / Listen Better moments
  • References specific patterns that showed up

✅ Comprehensive Length

  • 800-1200 words
  • Multiple sections with emoji headers
  • Call Quality Score

Example Output

================================================================================
✅ VOICE CALL REPORT TEST COMPLETE
================================================================================

📊 RESULTS:
   Status: success
   Call SID: SIM543D4BC8308D4C119F215B99E3E9F1
   Message ID: 2997

📝 REPORT PREVIEW:
--------------------------------------------------------------------------------
📊 CALL REPORT - 8 Minutes with Your Partner

🔍 WHAT HAPPENED
In this call, camily and their partner discussed frustrations around
household chores...

🧠 YOUR ATTACHMENT PATTERN: PURSUER
You're a Pursuer—when connection feels threatened, you pursue...

💪 MOMENTS YOU USED ASK BETTER
• "I feel alone in the responsibility." This statement expresses
  vulnerability...
--------------------------------------------------------------------------------

✅ Full report sent to GROUP conversation 10
================================================================================

Troubleshooting

"Conversation not found" error:

  • The simulated call was created but lookup failed
  • Check that the call_sid printed in Step 1 matches Step 2

"No transcripts found" error:

  • Transcript didn't create messages properly
  • Try --simple-parsing flag for simpler conversations

Report doesn't match Wren philosophy:

  • Check Langfuse staging has correct prompt version
  • Refresh prompt cache: Use langfuse-prompt-and-trace-debugger skill
  • Verify "production" label is set in Langfuse UI

Import error for test script:

  • Script location: api/src/scripts/test_voice_call_report.py
  • Make sure it exists and has correct imports

Related Skills

  • update-langfuse-staging-server-prompt - Update the voice report prompt
  • langfuse-prompt-and-trace-debugger - Debug prompt issues
  • run-voice-e2e - Full E2E voice pipeline test

Workflow for Prompt Changes

When updating the voice_call_summary_report prompt:

  1. Edit prompt in docs/cached_prompts/voice_call_summary_report_production.txt
  2. Push to Langfuse staging (use update-langfuse-staging-server-prompt skill)
  3. Test with this skill:
    docker compose exec api python src/scripts/test_voice_call_report.py \
  --transcript .claude/skills/test-voice-report/transcripts/yard_chores.txt
  4. Review report quality
  5. Iterate until satisfied
  6. Manually add "production" label in Langfuse UI

Quick Reference

# Test with existing transcript
docker compose exec api python src/scripts/test_voice_call_report.py \
  --transcript .claude/skills/test-voice-report/transcripts/yard_chores.txt \
  --duration 480

# Create new transcript (in arsenal - source of truth)
vim arsenal/dot-claude/skills/test-voice-report/transcripts/new_conversation.txt

# Sync to .claude
./arsenal/install.sh

# Test new transcript
docker compose exec api python src/scripts/test_voice_call_report.py \
  --transcript .claude/skills/test-voice-report/transcripts/new_conversation.txt \
  --duration 300