dsh-gray-test
Run the following command in DeepSeek Harness:
dsh plugin install LiLiLi-Zi/dsh-gray-test
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install LiLiLi-Zi/dsh-gray-test in the DeepSeek Harness terminal (source at https://github.com/LiLiLi-Zi/dsh-gray-test) and restart dsh to access the Gray Test panel from the Settings page.
About this plugin
When you rely on the DeepSeek official route through dsh, distinguishing a production model from a silent canary build is nearly impossible by eye alone. dsh-gray-test was built for exactly this: it spins up a configurable number of concurrent dsh sessions, streams each chain-of-thought token by token, and classifies every session in real time as gray, normal, or anomalous based on the ratio of I-m / I-will plus -ing verb occurrences to Let-me phrases.
The classification is deliberately tiered. Once a session shows the templated Let-me pattern the official V4 Pro uses for its CoT, the plugin cuts the connection early to save quota. Sessions flagged as gray are allowed to run to completion so you can later inspect the full output under ~/gray-test-runs/. Sessions that exhibit neither English template nor gray-style phrasing (for example a purely Chinese chain-of-thought) are logged as anomalous and also kept whole. You can also pick from dsh's built-in standard, code, minimal, or cordis agent presets, or load your own from .agent-presets, to mirror real workloads.
If you work daily with dsh on deepseek-v4-pro or deepseek-v4-flash and want a quick, repeatable answer to the question Am I on a canary model right now?, this plugin lets you fire off dozens of concurrent sessions and read the gray-model ratio in minutes rather than guessing one request at a time.
Use Cases
- Fire concurrent dsh sessions to check whether the current DeepSeek official route serves a canary build
- Stream-monitor each chain-of-thought to classify responses as gray, normal, or anomalous in real time
- Aggregate the gray-model ratio and archive full session outputs under ~/gray-test-runs/ for later review
Best For
- Developers who use dsh with deepseek-v4-pro or deepseek-v4-flash on a daily basis
- Users who need to confirm whether their account is routed to a production or canary model
- Engineers who want to batch-verify model behavior and avoid manual one-by-one testing
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.