open-cybernauts/open-range

AI Summary: This issue proposes adding a new 'abstract simulation mode' to the system for significantly faster reinforcement learning (RL) training. This mode will run episodes in microseconds, enabling rapid iteration and development by training in simulation and then validating on real Docker containers, all while maintaining the same API and observation types.

AI Summary: This issue proposes enhancing Non-Player Characters (NPCs) in a simulated corporate environment to behave like real employees by incorporating memory, reflection, and daily planning, inspired by Generative Agents research. The current NPC implementation is too simplistic, making their actions easily distinguishable from real user behavior, thus diminishing the value of stealth rewards for attackers.

AI Summary: This issue proposes adding deception capabilities for the Blue agent in a Red vs. Blue cybersecurity simulation. Blue would be able to deploy honeypots, canary credentials, and decoy services to detect and mislead the Red agent, enhancing the strategic depth of the simulation. This feature aims to mirror real-world SOC team tactics and create a more dynamic adversarial feedback loop.

AI Summary: This issue proposes tagging agent actions, golden paths, and Blue detections with MITRE ATT&CK technique IDs. This will make security training results comparable to threat models and increase the value of exported training data by allowing filtering by technique.

AI Summary: This issue proposes enhancing the cybersecurity training platform by introducing multiple interaction channels for Red agents to engage with NPCs, moving beyond simple shell commands. The goal is to simulate realistic social engineering tactics like phishing emails, voice calls, and document sharing, enabling Red agents to practice persuasion and Blue agents to detect sophisticated attacks.

AI Summary: This issue proposes the creation of a real-time episode dashboard for a simulation environment. Currently, users must parse JSON logs or grep container output to understand episode progression, which is inefficient for demos and debugging. The proposed solution involves a FastAPI backend with an SSE endpoint and a minimal frontend using HTMX and Mermaid.js/D3.js for visualization.

AI Summary: This issue proposes to improve the onboarding experience for new OpenRange users by making the offline demo more accessible and prominent. The goal is to allow newcomers to see the tool in action within 10 minutes, addressing a "cold-start" problem that hinders contribution and adoption.

AI Summary: This issue proposes adding a new `K8sBackend` implementation to support deploying OpenRange on real Kubernetes clusters, beyond the current local-only `KindBackend`. This will enable multi-tenant isolation and deployment on various cloud providers or bare-metal setups by leveraging standard `kubectl` commands.