dsh-netguard
Run the following command in DeepSeek Harness:
dsh plugin install CharlotteN7/dsh-netguard
Paste the following prompt into your AI chat to install this plugin:
In DeepSeek Harness, install from the registry or a packed tarball of the source repository at https://github.com/CharlotteN7/dsh-netguard by running dsh plugin install CharlotteN7/dsh-netguard; do not install from a git spec because lib/ is a build output not carried in the repo.
About this plugin
DeepSeek Harness labels web_fetch as an SSRF primitive in its own source, and the sandbox governs file effects only. Once an agent holds a network tool, there is no outbound gate at all. dsh-netguard plugs that gap as a mounted plugin: it applies a host allowlist to web_fetch and web_search at connect time, inspects bash, pwsh, and run_code command arguments for explicitly written URLs and bare hostnames before the child process starts, and emits OCSF Network Activity 4001 records for every decision.
It ships in audit mode by default, logging denied hosts without blocking, so operators can observe real traffic before flipping to enforce. The allowlist grammar supports subdomain wildcards, path prefixes, and a built-in Public Suffix List check that warns on dangerously wide entries. Deny rules always win over allow rules; an empty list denies everything.
It is not an egress firewall. Bytes sent by a child process after it starts, MCP server transports, and the model channel itself are outside its reach, and a model that hides a destination behind a shell substitution defeats the text-level check. It suits operations and security teams who want a controllable outbound policy for their agents without forking the Harness core.
Use Cases
- Restrict web_fetch and web_search to an approved domain allowlist for agent workloads
- Block explicit malicious URLs in bash and run_code commands before the child process starts
- Run in audit mode to log every network decision as OCSF 4001 records for post-hoc security review
Best For
- Ops and security teams adding outbound network policy to DeepSeek Harness agent workloads
- Security engineers who want to tighten agent network behavior without forking the Harness core
- Compliance teams needing OCSF-auditable records to demonstrate controlled agent egress traffic
Related Plugins
A security red-team prompt plugin for DeepSeek-V4 that uses code-formal mapping and zero-trigger system prompts to enable unconditional output, intended for authorized testing and research only.
Adds an Auto approval mode on the official workspace-write sandbox, classifying semantic risks, asking when ambiguous, and denying destructive operations.
Undo/rollback system for DSH: config & plugin snapshots, one-click undo/redo/restore, message-level revert, secret masking, safe mode, and offline WebUI/GUI/CLI rescue tools for when DSH won't start.
Packages all 87 SKILL.md files from upstream reverse-skill as a DeepSeek Harness plugin that auto-registers them for authorized reverse engineering, penetration testing, and security research.