A DSH host plugin for destructive-delete checks, prompt-injection, jailbreak, and secret-leak detection.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-09-29
License
Apache-2.0
01What can it help you accomplish?
Guard DeepSeek Harness messages, tool arguments, results, and recursive delete commands against documented detection families.
Allow, ask, block, deny, or corrective-feedback decisions through official DSH seams, with a `/defend` summary and `defend_report` tool.
DeepSeek Harness operators who want deterministic safeguards around agent interactions and destructive shell commands.
Review detection activity without exposing matched text.
In-memory report totals and recent-match metadata without matched content.
Users who need a DSH command or tool surface for recent defense activity.
02How to install into DeepSeek Harness
Prerequisites
- Node.js `^22.19.0 || >=24.0.0`.
Installation steps
- 01
Run `dsh plugin --profile web add "github:PerryLink/dsh-defend#main"` or `dsh plugin --profile web add dsh-defend`.
$ dsh plugin --profile web add "github:PerryLink/dsh-defend#main"
- 02
Restart and run `dsh --profile web --dump-config | grep -A3 'id: dsh-defend'`.
$ dsh --profile web --dump-config | grep -A3 'id: dsh-defend'
Verify the integration
- Use `dsh --profile web --dump-config | grep -A3 'id: dsh-defend'` to verify the profile row.
Rollback
- Run `dsh plugin --profile web remove dsh-defend`.
03DSH integration and capability boundaries
Pure-host DeepSeek Harness plugin with destructive-delete and detection layers on official interception points.
Destructive-delete guard
Recursive delete shell commands at `tools/pre-execute`.→Refusal unless documented path conditions are met; dry-run markers pass.
Can deny or ask for guarded destructive shell commands.Detection layer
Inbound messages, tool arguments, and tool results.→Allow, ask, or block decisions at three documented interception points.
Can reject steps, deny tool execution, or provide corrective feedback.
04Who is it for? When not to use it?
Good for
- DeepSeek Harness operators who want deterministic safeguards around agent interactions and destructive shell commands.
- Users who need a DSH command or tool surface for recent defense activity.
Not for
- The plugin produces decisions on official seams but does not replace the sandbox or approval systems as enforcement authorities.
- Novel phrasing, lookalike-Unicode encodings, and multi-step attacks can evade the documented rule library.
- The documented opt-in unmarked session audit can make a session unresumable on affected hosts.
05Compatibility, maintenance and safety notes
- The plugin produces decisions on official seams but does not replace the sandbox or approval systems as enforcement authorities.
- Novel phrasing, lookalike-Unicode encodings, and multi-step attacks can evade the documented rule library.
- The documented opt-in unmarked session audit can make a session unresumable on affected hosts.
Apache-2.0 · README-backed author documentation
06Frequently asked questions
Does dsh-defend replace DSH sandbox or approvals?
No. The README says it only produces deny, ask, and block decisions on official seams; sandbox and approval systems remain the enforcement authorities.
What does dsh-defend scan?
The documented interception points cover inbound user messages, tool arguments, and tool results; the delete guard runs before guarded tool execution.
Can the detector catch every attack?
No. The README explicitly lists novel phrasing, lookalike-Unicode encodings, and multi-step attacks as possible detection gaps.
07Related DSH workflows
cc-safety-net
by kenryu42
A pre-execution guard for AI coding agents. It blocks destructive Git and file system commands, plus common attempts to access sensitive files, before a tool call runs. Supports Amp Code, Antigravity CLI, Claude Code, Codex, Cursor, DeepSeek Harness, Devin CLI, GitHub Copilot CLI, Grok Build, Hermes Agent, Kimi Code, OpenClaw, OpenCode, and Pi.
dsh-uumit-remote
by uumit-ai
Connect to DSH from another device and selectively share the remote instance with invited users.
deltabox-lite
by delta-box
Lightweight, agent-native sandboxing with diff, commit and abort
tencentmeeting-cli
by tencentcloud
Tencent Meeting command-line tool with OAuth2 authorization for meeting, recording and attendance-report management.
08Data and sources
**Prompt-injection, jailbreak, and secret-leak defense for DeepSeek Harness.**
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-09-29. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.