dsh-where-am-i

DeepSeek Harness already provides session-level context such as the working directory, tool guidance, and deployment persona through its prompt and runtime-context layers. What it does not cover is the device-level picture: OS distribution, kernel release, CPU architecture, memory, uptime, timezone, shell/terminal, and Node version. dsh-where-am-i closes exactly that gap by injecting a single neofetch-style device snapshot as a durable user message at the first step of every brand-new session, so the model sees the host environment from the very start of the conversation.

The implementation deliberately uses a persistent user message rather than a system-prompt section. This guarantees that existing sessions keep a byte-identical system prompt, their request prefix and KV cache remain fully intact, and no later step of any session ever re-injects the snapshot. Data is sampled at injection time (so uptime and memory reflect the moment the session begins) and then lives as that session's own history until compaction. Ephemeral subagent workers are skipped by default to avoid repeating the snapshot on every fan-out, all values are read synchronously from node:os and /etc/os-release with zero subprocesses, architecture names follow the neofetch convention, and the fields option lets you emit only the lines you need.

The plugin is a natural fit for DeepSeek Harness users who want the model to reason about the host environment without manually pasting system info into every prompt: choosing build flags based on architecture, sizing generation parameters to available memory, handling timezone-aware scheduling, or simply giving the model the same at-a-glance context that neofetch provides in a terminal.