dsh-auto-approve

When running long tasks or frequently installing dependencies in DeepSeek Harness, the default workspace-write preset can severely disrupt your workflow with constant approval pop-ups for every sandbox upgrade, while switching to danger-full-access introduces significant security risks. dsh-auto-approve solves this dilemma by introducing an auto permission preset right in the middle, offering a low-friction yet secure automated approval experience.

The core capability of this plugin lies in its dual-validation mechanism, combining a deterministic danger list with an LLM classifier. For routine sandbox upgrade requests, it extracts the context and delegates the decision to the classifier for seamless auto-approval. However, if a request triggers high-risk regex patterns—such as destructive deletions or force pushes—or if the model expresses uncertainty, the approval flow safely falls back to a human prompt, ensuring critical operations remain under human supervision.

This plugin is perfectly suited for developers who need to run extended automated tasks, manage dependencies frequently, or develop plugins and presets. If you want to drastically reduce the tedious clicking of "Allow" while maintaining workspace sandbox boundaries and a complete audit trail, without assuming the severe risks of full-access mode, dsh-auto-approve is your ideal choice.