Show HN: I built Comrade – the security-focused AI agent
Inspired by the success of OpenClaw, I built Comrade, which is an open-source AI workspace for teams focused on security. It provides a premium interface for AI-powered workflows, built with transparency, extensibility, and local-first principles. Check it out here: https://github.com/LaurentiuGabriel/comrade Interesting. I believe security was the weakest link of OpenClaw, but how do you strengthen the security of your agent? Aren't they subject to prompt injection attacks by default due to the underlying LLM's that they're based on? I designed Comrade to have a "common sense" engine, where each plan/action goes through a filter before being suggested to the user to be approved. This is particularly enforced when the agent goes on a web page. It will always be aware of the source of the prompt, and if it's different from the Electron app where the user can interact with the agent, it will drop that instruction altogether.