AI Agent Hub
Back to plugins
dsh-plugin-rollout-scout preview

dsh-plugin-rollout-scout

Model Inference Updated 2026.08.26

Run the following command in DeepSeek Harness:

dsh plugin install SpookySandwich/dsh-plugin-rollout-scout

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

To install this plugin in DeepSeek Harness, run the command: dsh plugin install SpookySandwich/dsh-plugin-rollout-scout, or visit the GitHub repository: https://github.com/SpookySandwich/dsh-plugin-rollout-scout.

About this plugin

dsh-plugin-rollout-scout is a rollout scout tool designed for DeepSeek Harness users. Service providers often deploy new chat models via gray releases, leaving users unaware of which version they're assigned to, leading to inconsistent inference experiences or wasted tokens. This plugin automates detection to help users promptly identify new models and optimize resource usage.

Its core capability involves analyzing streaming chain-of-thought output, using heuristic scoring based on paragraph openings—such as 'Let me' indicating old models and 'I'll' at the start of the entire chain signaling a gray-release model. It also supports pre-stop mechanisms like generation speed limits and time-to-first-token thresholds to quickly abort old-model probes. The console interface offers real-time monitoring, score adjustments, and session management, making the detection process intuitive and controllable.

Ideal for developers, researchers, or power users of DSH who value inference efficiency and wish to automatically filter new models. The plugin reduces manual checking efforts, saves token overhead, and provides persistent retention and cleanup features for long-term use.

Screenshots

Use Cases

  • Automatically detect new models during gray releases by service providers
  • Monitor streaming chain-of-thought output to identify model types
  • Optimize inference token usage and avoid waste

Best For

  • DeepSeek Harness users
  • AI developers or researchers
  • Power users needing automated model detection