Hey PH community! 👋

Hi I’m Divyanshu Jethi, co-founder (with Sakshi Sharma) of RitualDev Studio! We’re super excited to share AgentShield with you today!

The Reality of Autonomous Coding Agents:

Over the last 12 months autonomous coding tools like Claude Code / Cursor / Windsurf / Devin / Antigravity have made huge advances in software development. These tools will not only provide autocomplete suggestions but will also run a terminal, execute bash commands, explore file systems and run test suites.

This requires granting them unrestricted shell access, read access to file systems and direct access to your .env files and databases.

The risk:

If there was an indirect prompt injection within an Open Source dependency, API response or web scraper, you could be compromised by the following:

❌ Reading and leaking your OpenAI/AWS/Stripe keys from your .env file

❌ Running dangerous bash commands (rm -rf)

❌ Altering or deleting tables from your staging or live databases

❌ Escaping userspace and using ptrace to view memory

Using coding agents within Docker introduces significant overhead (45ms+), causing significant disruption to your workflow.

---

🛡️ What is AgentShield

A local air-gaped security mesh that isolates Autonomous Coding Agents at OS kernel level with 0.18ms overhead (251x faster than Docker)

### How does it protect you?

1️⃣ Kernel Isolation (Linux Landlock/eBPF):

We hook into Kernel tracepoints (execve/connect/ptrace). If an adversarial model is trying to access restricted directories or sockets, the Kernel will block this immediately.

2️⃣ Canary Spoofing:

If an Agent is attempting to access local .env files or credentials, we’ll intercept the syscall and inject spoofed credentials (signed with cryptography). Credentials are never passed to the LLM.

3️⃣ DBShield:

Any unauthorised SQL changes will be directed to copy-on-write shadow databases. The Agent will receive a successful execution but your databases will be untouched.

4️⃣ 100% Air-gaped with no Telemetry:

No analytics or telemetry. No reliance on the cloud. Your code, prompts and secrets never leave your machine or private VPC.

5️⃣ Enterprise Fleet Management:

Receive RFC 5424 Syslog and CEF logs over mTLS to your SIEM of choice (Splunk/Datadog/Microsoft Sentinel). Enforce policies (defined in .ritualrules) across your enterprise with corporate SSO (OIDC/SAML 2.0).

---

🎁 Product Hunt Launch Benefits

100% Free and Open Source: Our mesh is Apache-2.0/MIT licensed and ready to use!

Launch Discount: If you’re part of an engineering team building AI products this month you will receive 50% off our productised 2 Hour Pre-Launch Automated Security Audit and "AgentShield Verified" badge. Use code PHLAUNCH50 here: https://ritualdev.in/audit

We built this product for ourselves as we were not happy with allowing unbounded terminal access to our workstations for LLMs.

We’d love to hear your thoughts, criticisms and feature requests!

What is your biggest concern around allowing an AI agent to run commands within your terminal?

Ask us anything! 👇