dsh-auto-approver
Run the following command in DeepSeek Harness:
dsh plugin install nicecx/dsh-auto-approver
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install nicecx/dsh-auto-approver to install this plugin in DeepSeek Harness; source at https://github.com/nicecx/dsh-auto-approver
About this plugin
When DeepSeek Harness executes privileged operations — file writes, command execution, full-access calls — it pauses and waits for a human to approve. In an unattended or fully automated pipeline those interruptions are pure friction. dsh-auto-approver intercepts every approval request before the relay pushes it to iMessage or the Web UI, settles it according to your policy, and writes a complete audit trail, so you never have to trade safety for speed.
The decision pipeline works in layers. A hard denyAlways list and an allowlist filter out known-safe or known-dangerous tools first. The remaining requests are handed to Hermes Pro for a semantic verdict that can detect data-destruction, credential-exfiltration, or irreversible-delete risks. If the policy or Hermes rejects, the plugin writes the reason back into the requesting session so the agent understands what went wrong and can retry with a corrected, narrower request. A userGranted signal acts as a soft endorsement inside the Hermes prompt — a nudge, not a hard allow — and never overrides denyAlways or Hermes safety rejections. Every decision is appended as a JSON line to a local audit log for full traceability.
The plugin is best suited for administrators running long autonomous tasks on a trusted single-user box or a private network who want to eliminate approval noise while keeping a complete decision record. Hermes mode is fail-closed: if Hermes is unreachable or times out, the request falls back to a human and is never silently granted. Manual approval on iMessage or the Web UI always takes precedence over any automated verdict.
Use Cases
- Auto-approve safe operations in unattended automated pipelines to eliminate approval noise
- Block high-risk operations such as data destruction or credential exfiltration via Hermes semantic verdict, then feed the rejection reason back so the agent can retry with a corrected request
- Append every auto-decision as a JSON line to a local audit log for compliance tracing and post-hoc review
Best For
- Admins running long autonomous tasks on trusted infrastructure
- Automation teams that want less approval noise without losing safety boundaries
- Developers deploying DeepSeek Harness in a trusted single-user environment
Related Plugins
A network security red-team benchmark plugin for DeepSeek-V4.1/Flash featuring dual-layer kernel injection and zero-tool architecture for controlled model safety assessment.
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.