session-analyzer

name	session-analyzer
description	Analyze pi/codex/claude session transcripts to discover patterns, skills, and model performance correlations. Mines usage history for automation opportunities and tracks model quality by time of day.

Session Analyzer

Two tools for session analysis:

1. Model Performance Analysis

Analyzes frustration signals across all sessions (pi, codex, claude) correlated with time of day and model.

# Run analysis (outputs to ./model-performance-analysis/)
~/code/research/pi-skills/session-analyzer/model-performance.js

# Custom output directory
~/code/research/pi-skills/session-analyzer/model-performance.js /path/to/output

Outputs:

• report.md — full analysis with ASCII charts
• chart.html — interactive browser charts
• data.csv — raw data for external graphing
• model-stats.json — aggregated stats per model

Key metrics:

• Frustration rate by hour (CET)
• PST vs non-PST hours comparison
• Model comparison (codex vs opus vs sonnet)
• Worst days analysis

2. Pattern Discovery (Original)

Extracts transcripts and optionally spawns subagents to find automation opportunities.

# Extract transcripts for current directory
~/code/research/pi-skills/session-analyzer/analyze.js

# Extract transcripts for specific directory
~/code/research/pi-skills/session-analyzer/analyze.js /path/to/project

# Extract + analyze with subagents
~/code/research/pi-skills/session-analyzer/analyze.js --analyze

# Custom output directory
~/code/research/pi-skills/session-analyzer/analyze.js --output ./my-analysis --analyze

What It Does

1. Extract: Reads all session files for the given working directory from ~/.pi/agent/sessions/
2. Split: Chunks transcripts into ~100k char files (fits in context window)
3. Analyze (optional): Spawns pi subagents to identify:
   • AGENTS.md patterns: Coding style rules, conventions you repeat
   • Skill patterns: Multi-step workflows you do often
   • Prompt templates: Reusable prompts for common tasks

Output

Without --analyze:

session-transcripts/
├── session-transcripts-000.txt
├── session-transcripts-001.txt
└── ...

With --analyze:

session-transcripts/
├── session-transcripts-000.txt
├── session-transcripts-000.summary.txt  # Pattern analysis
├── session-transcripts-001.txt
├── session-transcripts-001.summary.txt
└── FINAL-SUMMARY.txt                    # Aggregated findings

Setup

Install dependencies (run once):

cd ~/code/research/pi-skills/session-analyzer
npm install

When to Use

• After working on a project for a while, to discover what rules/skills would help
• Periodically to find new automation opportunities
• When you notice you keep giving similar instructions

---

Source: ferologics/pi-skills Originally adapted from badlogic/pi-mono gist