spy-on-session

Audit tool integrations in Claude Code sessions and generate structured findings.

4|Updated Jan 29, 2026

Search Tags:#debugging #audit #session #claude-code #tools #jsonl #findings

Ready to run

npx skills add https://github.com/danmestas/wardrobe --skill spy-on-session

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: spy-on-session
Source: https://github.com/danmestas/wardrobe/tree/main/skills/spy-on-session
Command: npx skills add https://github.com/danmestas/wardrobe --skill spy-on-session

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Audit and document how a tool integrates with a Claude Code session, fingerprinting changes, monitoring live transcripts, and organizing findings to identify bugs, inconveniences, and improvements.

Core Features & Use Cases

Fingerprint the before/after state of a session to attribute changes to the tool
Read and correlate the session JSONL transcript with tool behavior
Monitor live streams and hook outputs to validate integration correctness
Compile findings into a structured report for downstream triage

Quick Start

Open a Claude Code session on a project and run the spy workflow to capture fingerprints, JSONL, and findings.

Frequently Asked Questions about spy-on-session

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I audit tool integrations inside a Claude Code session?▼

To audit tool integrations inside a Claude Code session, you apply a spy workflow that captures pre/post bootstrap fingerprints, monitors live transcript JSONL files, and generates structured findings to identify integration bugs and improvements.

What is a bootstrap fingerprint when debugging Claude Code sessions?▼

A bootstrap fingerprint captures the before and after state of a Claude Code session to attribute specific state changes directly to the tool integrations, hooks, MCPs, and sidecars operating within that environment.

Can I monitor live JSONL transcripts to validate Claude Code hooks?▼

Yes, you can monitor live JSONL transcript streams and hook outputs to validate integration correctness, correlating the session transcript directly with tool behavior to evaluate end-to-end integration.

Do I need an operator session to debug MCPs and sidecars in Claude Code?▼

Yes, debugging MCPs and sidecars requires access to an active operator session, the live transcript JSONL file, and pre/post bootstrap fingerprints to accurately compare state changes and evaluate behavior.

What is the best way to identify bugs in Claude Code tool integrations?▼

The best way to identify bugs is to compile session monitoring data, live transcript correlations, and state fingerprints into a structured findings report, enabling downstream triage of integration inconveniences and errors.

Why are my Claude Code tool integrations causing unexpected state changes?▼

Unexpected state changes occur when tool integrations, hooks, or MCPs modify the session environment, requiring a pre/post bootstrap fingerprint comparison to accurately attribute the changes and generate structured findings for triage.