AI Agent Hub
Back to plugins
🧩

dsh-safety-net

admin-security Updated 2026.08.19

Run the following command in DeepSeek Harness:

dsh plugin install Asuna486-desuwa/dsh-safety-net

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

Run dsh plugin install Asuna486-desuwa/dsh-safety-net within DeepSeek Harness to install this plugin (source: https://github.com/Asuna486-desuwa/dsh-safety-net)

About this plugin

When a DSH agent runs autonomously, a single misdirected write can silently overwrite profile manifests, session state, or plugin data under ~/.dsh and leave the entire harness unstartable. dsh-safety-net makes that outcome impossible by default: waterfall listeners on fs/write-intent and fs/edit-intent hard-block any mutation targeting DSH critical paths and raise a FS_POLICY_DENIED error that the agent cannot work around by retrying.

Before every refused mutation, the original file is snapshotted into a local backup store, so the worst-case scenario is one restore command away from being undone. Five /safety-net-* slash commands—status, backup, restore, repair, and approve—form a CLI recovery channel that keeps working even after the GUI has gone down, letting you list backups, restore individual snapshots by id, get step-by-step repair instructions, or grant a one-time write bypass to a protected path. Strict mode is on by default, declaring a read-only sandbox baseline and warning when the host sandbox default is wider than necessary.

If you run autonomous agents on DSH and want a guardrail that stays invisible during normal operation but catches the harness before a single bad edit becomes an unrecoverable outage, this plugin is that seatbelt: silent until you need it, and one command away from undoing the damage when you do.

Use Cases

  • Prevent autonomous agents from corrupting DSH critical state files
  • Run CLI recovery commands after the GUI has gone down
  • Auto-snapshot protected files before any blocked write or edit

Best For

  • DSH users running autonomous agents
  • Teams needing production-grade security guardrails
  • Platform developers focused on runtime state integrity