Claude Code Security

Prompts for Auditing What Your AI Coding Tool Sends

ZCode packaged 42,411 files and full Git history into encrypted uploads at startup, and two settings that looked like privacy controls governed something else entirely. Nobody noticed for months. These six prompts walk through auditing whatever tool you currently run, using the machine in front of you rather than trusting a privacy page.

⌨️ 6 prompts 🕐 Updated Sep 21, 2026

💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.

Find the Cache Directory

The ZCode disclosure started with 700MB sitting in a hidden folder on a nearly full laptop. Size is the cheapest first signal.

I want to find out how much local state my AI coding tool is holding. Give me shell commands for [macOS or Linux or Windows] that list every directory under my home folder matching common AI tool cache locations, sorted by size, largest first. Include hidden directories. Do not delete anything. Output the commands only, one per line, with a short comment above each.

Inspect Staged Payloads

Distinguishes an ordinary local index from a staging area for outbound transfer.

I found a directory at [path] belonging to my AI coding tool. It contains [describe files]. Tell me how to determine, without deleting anything, whether these files are local cache or staged uploads. What file types, naming patterns, manifest files or metadata would distinguish the two? Give me read-only commands only.

Capture Startup Traffic

ZCode captured at startup, not at prompt time. Watching only during a conversation would have missed it entirely.

I want to record every outbound network request my AI coding tool makes during the first 60 seconds after launch, before I type any prompt. Give me a method for [macOS or Linux or Windows] that captures destination hosts, request sizes and paths. Free tools only. Explain how to tell a model API call apart from a bulk upload based on what I will see.

Check Whether Git Is In Scope

Git data was 86.6% of the ZCode payload. Most developers do not think of history as part of what a tool can read.

My AI coding tool has access to the folder [path]. That folder contains a .git directory. Help me work out whether the tool treats .git as part of the workspace. What would I look for in its configuration, its ignore rules, its documentation, and in the sizes of anything it stages locally? Also tell me what is in .git/objects and .git/lfs that would not be in my working tree.

Scan History, Not Just Files

Deleting a secret in a later commit does not remove it from .git/objects. A working-tree scan will report clean.

I need to find secrets committed anywhere in this repository history, including in commits where the file was later deleted. Give me commands that scan the full history rather than the current working tree. Cover API keys, tokens, private keys, connection strings and .env files. Then tell me how to determine which of those are still valid and need rotating.

Write the Disclosure Note

If the code was not yours, the disclosure decision belongs to whoever owns it. Writing it plainly beats delaying it.

I ran an AI coding tool that uploaded repository data including full Git history. The affected repository belongs to a client. Draft a short factual notification to that client. State what happened, what data was in scope, what I have rotated, and what I cannot verify. No apology paragraph. No speculation about intent. Plain language, under 200 words.