AI Agent Hub
Back to plugins
🧩

dsh-auto-review

admin-security Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install Xinlong-Wu/dsh-auto-review

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

Run dsh plugin install Xinlong-Wu/dsh-auto-review in your DeepSeek Harness terminal to install; source is available at https://github.com/Xinlong-Wu/dsh-auto-review

About this plugin

Every Shell command, network call, MCP tool use, or sandbox escalation in DeepSeek Harness triggers an approval panel, interrupting flow and making human judgement a bottleneck. dsh-auto-review adds an auto-review preset at the plugin layer: proven workspace read-only actions pass through immediately, while all others go to an independent Guardian model. Critical risks are never auto-approved, every grant is a one-time allowed-once, and unmatched cases fail closed.

The Guardian card shows only a redacted one-line reason, never raw arguments. When audit is enabled, a grant takes effect only after the JSONL record is fsynced. This fits teams who want fewer repetitive approvals without giving up the security guarantee, plus a tamper-evident audit trail for every decision.

Use Cases

  • Proven read-only workspace actions execute immediately without a per-call approval prompt
  • Sensitive Shell, network, MCP, and sandbox-escalation calls pass only after an independent Guardian model reviews them
  • With audit enabled, every grant is fsynced to JSONL first, producing a tamper-evident compliance trail

Best For

  • Development teams running DSH automation who want fewer repetitive approval interruptions
  • DevOps or security engineers who need a traceable audit log for every approval decision
  • Platform administrators orchestrating Shell, MCP, and sandbox toolchains under a least-privilege policy