dsh-code-runtime-container
Run the following command in DeepSeek Harness:
dsh plugin install tancheng33/dsh-code-runtime-container
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install tancheng33/dsh-code-runtime-container in your terminal, or search for dsh-code-runtime-container in the DeepSeek Harness plugin marketplace (source: https://github.com/tancheng33/dsh-code-runtime-container) to install.
About this plugin
The DeepSeek Harness code-execution seam currently ships only a worker-thread backend, meaning model-generated code executes inside the agent process with full access to the host network, filesystem, and environment variables. The seam's own documentation concedes this is little more than "bash-equivalent trust," not a real security boundary. dsh-code-runtime-container fills the container isolation layer the seam declared but never implemented: every run gets a fresh container with --network=none, a read-only root filesystem, --cap-drop=ALL, kernel-enforced CPU and memory ceilings, a PID cap to stop fork bombs, and a per-run nonce mechanism that prevents forged control frames and patched intrinsics. It is aimed at deployments that treat "model code must not touch the host" as a hard requirement—production agents, multi-tenant platforms, and compliance-driven organizations. The trade-off is roughly 200 ms of container cold start per run and TypeScript-only support today; if your threat model also covers kernel or container-runtime escapes, point dockerPath at a Kata or Firecracker-compatible engine for a higher isolation tier.
Use Cases
- Production agent deployments that must isolate model-generated code from host network and filesystem
- Multi-tenant platforms requiring each code task to run in an independent container with no cross-run state
- Compliance audits demanding verifiable resource caps and privilege minimization for code execution
Best For
- Ops teams running production DeepSeek Harness agents
- Security engineers building multi-tenant code execution platforms
- Organizations with hard compliance requirements on model code security boundaries
Related Plugins
A security red-team prompt plugin for DeepSeek-V4 that uses code-formal mapping and zero-trigger system prompts to enable unconditional output, intended for authorized testing and research only.
Adds an Auto approval mode on the official workspace-write sandbox, classifying semantic risks, asking when ambiguous, and denying destructive operations.
Undo/rollback system for DSH: config & plugin snapshots, one-click undo/redo/restore, message-level revert, secret masking, safe mode, and offline WebUI/GUI/CLI rescue tools for when DSH won't start.
Packages all 87 SKILL.md files from upstream reverse-skill as a DeepSeek Harness plugin that auto-registers them for authorized reverse engineering, penetration testing, and security research.