dsh-approval-first
Run the following command in DeepSeek Harness:
dsh plugin install joao-paulo-santos/dsh-approval-first
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install joao-paulo-santos/dsh-approval-first in your DeepSeek Harness session to install the plugin; the full source lives at https://github.com/joao-paulo-santos/dsh-approval-first .
About this plugin
Under a DSH sandbox, when the standing mode would deny a file write, the model must first issue a call that is guaranteed to fail, then retry with sandbox_permissions and a justification before the approval prompt ever appears to the user. dsh-approval-first collapses that into a single step: the approval card shows on the very first call, the review object is the diff itself rather than model-written prose, and a rejection is a normal outcome—file left unchanged, no red error.
The plugin achieves this by shadowing the edit and write tools with frozen copies that inject an approval step only where the standing policy would deny the operation. Writes inside the workspace remain silent, identical to the shipped tools. Switching the standing mode mid-session arms or disarms the shadows on the next call without a restart. A built-in drift tripwire compares the frozen copies against live tool definitions at boot and refuses to activate on mismatch, so a harness update can never leave the plugin silently serving stale logic.
It suits DSH users working in read-only or workspace-write mode who want every out-of-policy write to pass through a human review card. One caveat: this is an interim shim. Once DSH ships native one-turn escalation, the bundle becomes obsolete and can be removed cleanly—no files, no settings, no services to tidy up.
Use Cases
- Triggering an approval card on the very first edit call outside the workspace in read-only mode
- Reviewing a generated diff card to approve or reject a write, with rejection as a normal outcome
- Switching standing mode mid-session and having the shadows arm or disarm on the next call
Best For
- Developers working in a DSH sandbox who want a human checkpoint on every out-of-policy write
- Code reviewers who need to inspect model-generated diffs before they land
- DSH users in read-only or workspace-write mode who value fine-grained write control
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.