AI Agent Hub
Back to plugins
🧩

dsh-codex-approval

admin-security Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install 040822/dsh-codex-approval

Paste the following prompt into your AI chat to install this plugin:

To install this plugin in DeepSeek Harness, please use the command dsh plugin install 040822/dsh-codex-approval, or visit the source code at https://github.com/040822/dsh-codex-approval.

About this plugin

DeepSeek Harness provides strong sandbox isolation, but its approval strategies are somewhat "all-or-nothing". Native strategies rely on static permission modes or completely hand over confirmation to humans, lacking intelligent judgment and risk assessment for specific commands. To fill this gap, the dsh-codex-approval plugin was created to bring an intelligent experience similar to Codex CLI to dsh.

By introducing command-level rules and AI risk assessment, this plugin builds a complete automated approval decision chain. You can use glob rules (like Bash(git *)) to define automation policies or let the AI output risk levels (low/medium/high) and authorization suggestions in real-time based on command content. It supports three flexible approval modes, from the completely bypassed "Manual Mode" to the fully automated "AI-Auto Mode", allowing machines and humans to take turns in the approval process.

Whether you are a developer seeking ultimate efficiency or an operations professional needing to maintain a safety baseline in automated scripts, this plugin is the ideal companion. It not only drastically reduces useless human pop-ups but also blocks high-risk operations at critical moments, making DeepSeek Harness's automation capabilities safer and more reliable.

Use Cases

  • Intelligent control of sensitive operations in automation scripts
  • Risk assessment before executing AI-generated code
  • Fine-grained approval strategies for multi-level commands

Best For

  • Developers balancing high efficiency and security
  • Engineers responsible for automated operations
  • Teams needing command-level fine-grained permission control