How does this skill detect user frustration?

It extracts patterns from user records where tools are denied, requests are interrupted, or direct corrective language like 'wrong' or 'stop' is used.

Where are the Claude Code session transcripts located?

Transcripts are stored locally under ~/.claude/projects/ in project-specific directories named after the project path with hyphens replacing slashes.

Where are the analysis results saved?

Findings are exported to the .planning/kaizen/ directory as structured markdown reports containing evidence, severity levels, and specific remediation steps.

Can this skill help reduce agent errors?

Yes, it specifically identifies patterns such as 'Edit-before-Read' errors and stale edit targets to help you optimize agent prompts and automated hooks.

What type of data does the Transcript Analysis skill process?

It analyzes JSONL files containing assistant tool calls, user inputs, system metadata events, and hook execution logs to map the agent's decision-making process.

Claude Code Transcript Analysis

Name: Claude Code Transcript Analysis
Author: Jamie-BitFlight

byJamie-BitFlight

•

Analytics & Monitoring

Analyzes session transcripts to identify performance bottlenecks, tool misuse, and workflow improvement opportunities using data-driven forensics.

The Transcript Analysis skill provides a comprehensive framework for auditing and optimizing Claude Code behavior by mining JSONL session data. It leverages DuckDB SQL patterns and PM4Py process mining methodology to detect 10 critical dimensions of agent performance, including user frustration signals, tool-sequence anti-patterns, and systemic red herrings. This skill is indispensable for developers looking to perform deep session forensics, refine agent prompts, and automate repetitive manual workflows through data-backed kaizen analysis.

Key Features

01DuckDB-powered SQL querying for structured JSONL transcript forensics

02User frustration signal extraction from session interaction logs

03Automated recommendation engine for pre-tool hooks and prompt patches

04Automated tool misuse detection and anti-pattern identification

0526 GitHub stars

06Advanced process mining for tool-sequence and shortest-path analysis

Use Cases

01Debugging complex agent behavior and infinite tool-call loops

02Identifying candidates for automated PreToolUse or SubagentStart hooks

03Performing session forensics to discover why an agent failed a specific task

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jamie-bitflight/claude_skills transcript-analysis

For use in Claude.ai and ChatGPT

Download Skill