NanoClaw Partners with Echo to Patch Known Vulnerabilities in the Agent's Runtime, Browser, and Libraries

Protect your agent, not just yourself from your agent.

Isolation is powerful, and NanoClaw has leaned on it: every agent runs in its own sandbox, your credentials sit behind a policy-enforcing gateway that injects them into network requests after they leave the sandbox, and a human stays in the loop on anything sensitive. That contains what a compromised agent can do.

What it doesn’t do is change the software inside the box. To browse the web or open a file, your agent works inside its own full computer: a browser, parsers, an npm dependency tree, each with its own CVE history. Isolation limits the blast radius if the agent is compromised; it doesn’t stop an attacker from targeting the agent itself. Attackers can compromise the agent’s environment to exfiltrate data or use the agent’s gateway to take actions. That’s the thing most people never look at: the software the agent uses and runs on.

NanoClaw is partnering with Echo to give the community a hardened agent runtime: the same NanoClaw environment, with the software inside rebuilt by Echo. It’s live now, and it makes NanoClaw the first open-source agent framework whose runtime ships with near zero known CVEs.

Isolation stops at the sandbox wall

Software inside sandboxes carries thousands of known vulnerabilities, and most of them sit unpatched. Don’t take our word for it: point Trivy or Grype at OpenClaw, unhardened NanoClaw, or Hermes and you’ll see hundreds to thousands of known vulns — Chromium alone, 35+ million lines of code, accounts for a steady share. Not because anyone decided they were safe, but because exploiting them used to take a serious attacker, and most systems were never worth the effort. Agents change that math by making those attacks cheap to automate.

Frontier models and agents are robust against the “Ignore all previous instructions” type of prompt injections — see Anthropic’s and OpenAI’s recent model cards and research. A modern prompt injection is like a social engineering attack: it lures the agent to an innocent-looking page, and a known hole in the browser turns that visit into a compromise. Or it hands the agent a file, and a vulnerable parser turns opening it into remote code execution. Once an agent is compromised, an attacker can read its memory files and its workspace, and send requests through every credential the gateway will inject for it.

You keep an employee’s laptop and browser patched. Your agent’s environment deserves the same.

How the hardened runtime works

NanoClaw and Echo built it together. NanoClaw defines the agent environment and the orchestration layer that handles isolation and policy enforcement around it. Echo rebuilds the software inside the agent’s runtime environment from scratch, with only the essential components — and patches what remains, rather than just removing what isn’t needed — cutting the known-vulnerability count from thousands to near zero.

The result is a full agent runtime with all the things an agent needs baked in: Chromium, Node, Bun, pnpm, Corepack, git, curl, and unzip. It’s fully compatible with NanoClaw’s requirements, minus the security holes.

And it stays secure

A patched environment is only useful if it stays patched. When a new vulnerability is discovered anywhere in the world, Echo’s agents research it, identify the affected software, find or develop a fix from upstream commits, security advisories, and mailing lists, run compatibility testing, and open a pull request for human review.

How you turn it on

The hardened runtime is free, and it’s opt-in: turning it on just requires one-time authentication with NanoClaw (so we can protect ourselves from abuse). New setups can choose it during install, and existing users can enable it with /update-nanoclaw. See the docs for details.

Don’t want it? Nothing breaks. The unauthenticated path of building your own agent runtime locally stays exactly where it was.

The hardened runtime is there when you want it, not something imposed on you.

Why this matters

I’ve argued that agents should not be trusted. Security should come from things you can verify, not instructions or heuristics. But once agents start to do real work, handle sensitive data, take actions (with your approval), you need to protect your agent, not just protect yourself from your agent. It’s not enough to say “your agent is walled off if something goes wrong.” The software it runs on, the browser it surfs the web with and the libraries it uses shouldn’t carry known security holes.

NanoClaw contains the agent, Echo hardens the ground it stands on and the tools it uses.

The hardened NanoClaw runtime is available to the community today.

---

NanoClaw is an open-source, secure runtime and orchestration layer for agent teams. Star it on GitHub.