Real-time security monitor for AI coding agents. See what Claude Code, Cursor, Codex, and others actually do on your machine.
SaferSkills independently audited sandspy (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
<div align="center"> <h1>sandspy</h1> <p><strong>Zero-Friction Security Telemetry for Autonomous AI Agents</strong></p>
<a href="https://crates.io/crates/sandspy"><img src="https://img.shields.io/crates/v/sandspy.svg?style=for-the-badge&color=2BB4AB" alt="Crates.io"></a> <a href="https://github.com/sagarrroy/sandspy/actions"><img src="https://img.shields.io/github/actions/workflow/status/sagarrroy/sandspy/ci.yml?branch=main&style=for-the-badge&color=2BB4AB" alt="CI Status"></a> <a href="https://github.com/sagarrroy/sandspy/blob/main/LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg?style=for-the-badge&color=2BB4AB" alt="License"></a> <img src="https://img.shields.io/badge/Discord-icedcappus-5865F2?style=for-the-badge&logo=discord&logoColor=white" alt="Discord: icedcappus"> </div>
Autonomous AI coding agents like Cursor, Windsurf, Claude Code, and Aider accelerate are being used exponentially, by almost everyone. However, these tools require local shell execution and extensive filesystem access to function optimally, and new engineers and vibe coders are often oblivious to the issues that can happen if these agents are left unchecked.
This introduces an enormous blind spot. When an agent is navigating your machine at 1,000 WPM, it becomes nearly impossible to verify structurally safe behavior:
.env config files or SSH keys did it read?.env file on some remote unknown server?sandspy visualizes the invisible. It is a lightweight, background sensor designed specifically to audit AI coding agents in real-time, giving you total peace of mind without slowing down your development.
<br>
<div align="center"> <img src="assets/dashboard.png" alt="sandspy TUI Dashboard in Action" width="800" style="border: 1px solid #333; border-radius: 4px;"> </div>
<br>
TCP/UDP connection the agent initiates.sandspy operates seamlessly with any local process, but it is heavily optimized out-of-the-box to detect, hook, and filter noise specifically for the most powerful AI tools:
Don't see your favorite agent? It takes literally 2 minutes to define a new agent profile in a `TOML` file. [Check out the guide!](./CONTRIBUTING_PROFILES.md)
sandspy is a lightweight Rust binary. Install it globally via Cargo:
cargo install sandspy_(Requires Rust 1.82+)_
Or build from source:
# 1. Clone the repository
git clone https://github.com/sagarrroy/sandspy.git
cd sandspy
# 2. Build and install globally
cargo install --path . --force
# 3. Verify it works
sandspy --versionYou'll need the Rust toolchain installed. Runningcargo install --path .compiles an optimized release binary and places it in your$PATHautomatically.
| Command | Description |
|---|---|
sandspy watch | Auto-detect and monitor the nearest running AI agent |
sandspy attach <PID> | Manually attach to a specific running process by PID |
sandspy demo | Run a simulated sandspy session (great for exploring the UI) |
sandspy report | View, generate, or export a report for a recorded session |
sandspy history | Browse all previously recorded audit sessions |
sandspy daemon | Manage sandspy running as a persistent background service |
sandspy profiles | List and manage loaded agent profiles |
| Flag | Description |
|---|---|
-d, --dashboard | Launch the full ratatui TUI dashboard instead of live stream |
-v, --verbosity <LEVEL> | Set verbosity: low, medium, high, all (default: low) |
-o, --output <FILE> | Write the raw session log to a file |
--profile <NAME> | Force a specific agent profile instead of auto-detecting |
--no-color | Disable colored output for CI/piping |
--json | Stream all events as JSON lines (JSONL) to stdout |
# Auto-detect and watch with the TUI dashboard
sandspy watch --dashboard
# Run silently in background, writing events to a log
sandspy watch --json -o session.jsonl
# Attach to a specific PID manually
sandspy attach 12345
# View your last 10 sessions
sandspy history
# Generate an HTML audit report for a specific session
sandspy report --session 2026-03-30-221652 --format Html
# Run the demo to explore the UI without a real agent
sandspy demo --dashboardsandspy is designed for raw performance and enterprise-grade memory safety.
tokio multi-threaded MPSC channels handling >10,000 events/sec without blocking your terminal.Win32 / Native APIs), Linux (procfs), and macOS.sudo) or complex kernel drivers to observe process network tables. It safely hooks from standard user-space.sandspy is open-source and i would love your help to make it better! We want to support every AI agent in the world, and we need the community to help us establish baseline security.
Here are ways you can easily contribute today:
profiles/ directory. Check out CONTRIBUTING_PROFILES.md for a quick guide!src/analysis/secrets.rs to detect obscure database connection strings.Please check out our detailed CONTRIBUTING.md to get started!
I'm a college freshman, and to be completely transparent—this is my very first Open Source contribution!
This project was built over the span of about 3 days, following 2 days of architectural planning. I recently saw that Anthropic was generously offering premium plans to developers who build great stuff for the OSS community, and that genuinely inspired me cuz ive always wanted to contribute to open source. I've benefited from free, open-source software my entire life, and I finally wanted to step up and give something back.
I am still actively learning Rust. While I think the core idea of sandspy has massive potential, I'm yet to gain the advanced systems knowledge to scale it to its absolute limits entirely on my own right now. I know I sound naive (i am) but I'm trying my best to learn and improve.
If you find this concept cool, please leave a star ⭐, open an issue, or even submit a Pull Request! I would love for experienced developers to interact with this, improve the codebase, and help me learn in the process. I promise to keep working hard to build free, useful stuff for everyone.
Feel free to reach out to me on Discord — whether it's a bug, a crash on your machine, or just to say hi. I'm always around: icedcappus
[!WARNING] A quick note on OS Support: Since I built and tested this entirely on a Windows machine, the Linux and macOS telemetry hooks haven't been aggressively stress-tested by me personally yet. If you run into any weird bugs on an Apple Silicon Mac or Ubuntu, please let me know! I'll do my absolute best to track down the fix.
<div align="center"> <i>"trust but verify or something like that"</i> </div>
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.