Agent Isolation and Sandbox Security at Warp
Warp combines a native GPU-rendered terminal engine with autonomous agent orchestrators like Claude Code, Gemini CLI, and custom agents. Running autonomous coding agents inside developer environments requires hard execution boundaries. Security is non-negotiable. Zero escapes tolerated. Without sandboxing, rogue agent tools could leak SSH keys or execute destructive shell commands. Warp isolates agent processes using platform-native controls: seccomp profiles, Landlock, and macOS seatbelt sandbox policies. We enforce strict boundaries. The desktop client connects to the cloud orchestrator through multiplexed Tokio channels with mTLS. All tool calls route through an immutable authorization broker written in Rust. You will design the kernel-level isolation layers and credential vaults that prevent unauthorized file system writes during agent refactoring sessions.
Target base compensation ranges from $230,000 to $290,000 USD per year, with significant startup equity. Equity included.
Core Engineering Responsibilities
- Implement fine-grained capability sandboxes for local agent tool invocations using Linux cgroups and macOS sandbox APIs.
- Architect an immutable audit log daemon in Rust capturing all filesystem and network IO initiated by AI workers.
- Build cryptographically secure IPC channels connecting the native client UI to background orchestrator processes.
- Integrate enterprise SSO, role-based tool restrictions, and secret scrubbing into the developer terminal flow.
Technical Requirements
- Demonstrated experience shipping production systems in Rust, C, or modern C++.
- Deep understanding of Linux security primitives, sandbox APIs, and POSIX process lifecycles. Rust fluency expected.
- Familiarity with asynchronous networking using Tokio and cross-platform IPC patterns.
- Disclosed compensation: $230,000 to $290,000 USD base salary plus equity.
