Prompt injection is the #1 OWASP LLM risk — your agent has no defense

Thu, 16 Jul 2026 18:18:00 +0000

Risk: prompt injection
Prompt injection can turn your agent into an attack vector.

OWASP LLM01:2025 ranks prompt injection as the #1 risk for LLM applications. An injection hidden in a dependency README, a tool output, or an MCP server response can instruct your agent to read secrets, run shell commands, or modify configs — all appearing as normal agent activity. Guard intercepts those actions before they execute and gives you a receipt for every blocked attempt.

Guard dashboard preview

No Guard installed

What Guard intercepts

Guard checks every tool call, file read, shell command, and MCP action against your policy before it runs.

•

Install Guard locally

•

Run one protected session

•

Review the first receipt

What this means

●

Blocks injected shell commands and secret reads before execution

●

Intercepts MCP tool calls and network calls that deviate from the task

●

Receipts show what was blocked, why, and what the agent was trying to do

Next action

Install Guard on the machine where your agent runs

Run one normal coding session

Open the receipt to see what Guard intercepted

Expected proof

The first receipt shows blocked actions with full context — what triggered the block and what data was at risk.

If nothing is blocked on the first run, the receipt still proves your session was monitored end-to-end.

Install Guard locally

If the button does not work, use this link:
https://hol.org/api/email/track-click?token=yEFOk9uKlB-Lg5cFAuaAOFxdfLl6WlHDXUu5VdVbxnU&scope=guard-lifecycle&target=https%3A%2F%2Fhol.org%2Fguard%2Factivate%3Futm_source%3Demail%26utm_medium%3Dlifecycle%26utm_campaign%3Dguard_nurture%26utm_content%3Dinstall_reminder_guard_customer%26guard_nurture_stage%3Dinstall_reminder

Need help? support@hashgraphonline.com

Unsubscribe from Guard lifecycle emails

Claude Code needs receipts

Thu, 25 Jun 2026 18:22:00 +0000

View Online

AI coding agents are getting more powerful every week.
Claude Code can read files, modify code, run commands, and execute multi-step tasks across your project.

That’s what makes it useful. But that’s also what makes visibility important.

As AI coding tools become more autonomous, understanding what happened becomes just as important as the result itself. That’s why we built HOL Guard.

Try HOL Guard Free

HOL Guard adds a lightweight safety layer for AI coding agents, providing:
• Approval workflows
• Action receipts
• Visibility into agent activity
• Support for Claude Code, Cursor, Codex, Gemini CLI, OpenCode, and more

No paid account required. No complicated setup.
Just install and start using it.

We’re actively looking for early users and feedback. If you’re using AI coding agents today, we’d love to hear what you think.

Safe building,
The HOL Team

Andris Veliks - HOL Guard

Prompt injection is the #1 OWASP LLM risk — your agent has no defense

Claude Code needs receipts