Sandbox & Runtimedsh-plugin

dsh-perm-guard

Unofficial community plugin for DeepSeek Harness web that adds a middle approval tier between workspace-write and danger-full-access: safe operations in trusted directories are auto-approved, while destructive operations always prompt for human confirmation. Includes Standard/Aggressive modes, 11 per-category switches, an in-plugin audit trail, and a persisted config at ~/.dsh/perm-guard.json.

Stars
1
Forks
3
Open issues
0
Last push
Sep 29, 2026
Latest release
v0.4.2
24h Growth
+0
7d Growth
+0

Installation

dsh plugin --profile web add "github:a903067276-rgb/dsh-perm-guard#main"

Overview

Unofficial community plugin for DeepSeek Harness web that adds a middle approval tier between workspace-write and danger-full-access: safe operations in trusted directories are auto-approved, while destructive operations always prompt for human confirmation. Includes Standard/Aggressive modes, 11 per-category switches, an in-plugin audit trail, and a persisted config at ~/.dsh/perm-guard.json.

  • Two modes: Standard auto-approves inside trust directories; Aggressive is location-unrestricted but still prompts for destructive operations
  • 11 per-category tri-state switches (auto / ask / deny) with personal defaults
  • Dangerous operations always prompt: deletes, disk ops, privilege escalation, and curl|sh download-execute
  • Audit trail records each decision (approved / forwarded to human / rejected) with timestamp and command summary
  • Persistent config in ~/.dsh/perm-guard.json that survives restarts; zero host dependencies
  • Auto button in the composer tool row toggles auto-approval; turning it off restores the host's default approval behavior
  • Interception before the host prompt: safe operations answered allowed-once; unparseable commands always fall back to asking
  • Command-level firewall via tools/pre-execute intercepts dangerous categories before the sandbox rejects them

Related plugins