AI Agent Hub
Back to plugins
dsh-turn-approval preview

dsh-turn-approval

admin-security Updated 2026.08.15

Run the following command in DeepSeek Harness:

dsh plugin install arrow949/dsh-turn-approval

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

Run dsh plugin install arrow949/dsh-turn-approval in your terminal (source: https://github.com/arrow949/dsh-turn-approval), then restart DSH Web and hard-refresh your browser (Ctrl+Shift+R) to load the plugin.

About this plugin

By default, DSH approval cards offer only two choices when an Agent requests danger-full-access: deny, or allow once. When a single task needs several high-privilege operations in a row, clicking "allow" repeatedly is disruptive and invites careless approval.

dsh-turn-approval adds a third button, "Allow for this task": once you click it, all subsequent danger-full-access requests within the current turn are auto-approved, and the grant expires the moment the turn ends. The authorization lives in memory only—nothing is written to disk, it never crosses session boundaries, and child agents do not inherit it. A process restart or plugin uninstall resets everything. The plugin renders through the official conversation.composer slot and modifies no DSH bundle files; uninstalling restores the two-button default immediately.

Ideal for DSH Web users who frequently trigger high-privilege approvals and want to cut repetitive clicks without giving up the per-request safety guardrails. Pure ESM JavaScript, zero extra dependencies, MIT licensed.

Screenshots

Use Cases

  • Agent fires multiple high-privilege requests within a single task, eliminating repetitive approvals
  • Grant temporary access for the current turn only, with automatic expiry on turn end
  • Need an approval granularity between 'allow once' and 'open the whole session'

Best For

  • DSH Web users who frequently encounter danger-full-access prompts
  • Security-conscious users who want fewer clicks without giving up per-action control
  • Developers running multi-step Agent tasks in DSH Web with a focus on permission boundaries