dsh-clawrouter: A Safety Review DSH plugin for DeepSeek Harness

blockrunai/dsh-clawrouter

Skills & AgentsVerified
Listed on 2026-08-20
Page last updated 2026-08-20

A DSH plugin that reviews dangerous tool calls before execution, with 67 models from one wallet paid per request.

Let a stronger model review and decide before dangerous commands run, offering a third option for DeepSeek Harness's Full Access mode.

install
dsh plugin --profile web add github:blockrunai/dsh-clawrouter
Category
Skills & Agents
Platform
DSH-Plugin
Author
blockrunai
Distribution
Plugin

dsh-clawrouter Key Features

Pre-execution safety review67 models one walletPay-per-request no API keysEnforced by real executor

dsh-clawrouter Repository

blockrunai
Publisher
blockrunai
Repository
blockrunai/dsh-clawrouter
19
Stars
1
Forks
0
Watchers
2
Open issues
LanguageTypeScript
LicenseMIT License
Website
Repo last push2026-08-17 06:32:05
Repo created2026-08-14 08:58:39
Topics
ai-agentsblockruncode-reviewdeepseek-harnessdshdsh-pluginllm-routerx402
Default branchmain

dsh-clawrouter Repo Summary

dsh-clawrouter is a safety review plugin (DSH plugin) for DeepSeek Harness, maintained by BlockRunAI, MIT licensed, and written in TypeScript. It addresses the core problem: in Full Access mode, an agent might execute dangerous commands (like accidentally deleting your home directory), while manually approving every command is tedious. This plugin introduces a stronger review model (e.g., Claude) that reads the dangerous tool call before it runs and returns allow / deny / ask, enforced by the real tool executor, not just by a prompt. It also registers a BlockRun provider route, making the reviewer and 67 other models reachable from one wallet, paid per request over x402 (USDC), with no accounts, no API keys, and no credit card.

What core safety review mechanism does the dsh-clawrouter DSH plugin provide?

When the agent proposes a dangerous command like rm -rf ~, dsh-clawrouter first has a stronger model (e.g., Claude) review it and return one of three verdicts: allow, deny, or ask. This verdict is not a soft constraint in the prompt but a hard gate enforced by the real tool executor, intercepting risk before the command actually runs. Compared to Harness's default Full Access all-or-nothing mode, it offers a third option: neither manually approving every command nor completely letting go.

How does dsh-clawrouter integrate 67 models and x402 payments?

The plugin registers a BlockRun provider route, allowing access to the reviewer model and 67 other models through a single wallet, paid per request via the x402 protocol using USDC. The whole process requires no account creation, no API keys, and no credit card, significantly lowering the barrier to multi-model usage.

How do I install dsh-clawrouter?

Use the DSH plugin manager with the following command:

bash
dsh plugin --profile web add github:BlockRunAI/dsh-clawrouter

After installation, the plugin automatically registers the review mechanism and the BlockRun provider route; no additional configuration is needed.

Which DeepSeek Harness versions and platforms does dsh-clawrouter support?

As an officially recognized DSH plugin, it depends on the @deepseek-ai/dsh-* packages (such as dsh-agent, dsh-tools, dsh-llm, etc.) and integrates deeply with the DeepSeek Harness plugin ecosystem. The project was created on 2026-08-14 and last updated on 2026-08-15. It is still in an early stage (9 stars and 1 fork on GitHub) but already offers complete review and payment capabilities.

View more DSH plugins at dsh-plugin.org

DSH-Plugin FAQ