dsh-env-profile
Run the following command in DeepSeek Harness:
dsh plugin install Yvesgao/dsh-env-profile
Paste the following prompt into your AI chat to install this plugin:
Install dsh-env-profile in DeepSeek Harness by running dsh plugin install Yvesgao/dsh-env-profile, with the source repository at https://github.com/Yvesgao/dsh-env-profile .
About this plugin
Every new session forces the model to rediscover the machine from scratch: which tools are installed, their versions, which paths are reliable, which network link is fast. That repeated probing is slow and burns tokens for no lasting gain. dsh-env-profile solves exactly this. As a DSH global-experience plugin, it runs a one-time host-side probe, distills the results into a cross-session profile, and hands the model a working knowledge of the environment from the very first turn so it stops re-running env, which, and ping round by round.
The closed loop has three stages. Probing fires in parallel across 33 common commands (existence plus version where available), key filesystem paths, a TCP-443 handshake timing, and a 1 MiB disk read/write sample—all deterministic host-side operations that cost zero tokens and are cached for a configurable interval (default 6 hours). Accumulation subscribes to session events, folds per-turn tool calls, errors, and token ledgers into per-session stats, and uses the session's own model every N turns to extract user preferences, project conventions, and useful paths as knowledge entries, deduplicated by normalized text and stored globally. Injection compresses the profile into a byte-deterministic summary (hard-capped at 600 characters by default) at system-prompt slot 500; when the profile is unchanged the summary is identical byte-for-byte every turn, so DeepSeek's prefix cache keeps hitting, and even when the summary does shift only the tail is invalidated while the large stable prefixes above remain cached.
It is aimed at developers who run DSH intensively and care about token efficiency: if you open a dozen sessions a day, keep asking the model to re-verify tool versions, paths, or network latency, or simply want the model to carry forward the machine's quirks across sessions, this plugin gives every new conversation the full environment context up front and redirects the saved probing turns and explanatory tokens back to the actual work. The current skeleton targets Windows; POSIX support is already on the roadmap.
Use Cases
- Onboard every new session with an instant snapshot of local tools, paths, network, and disk I/O
- Fold per-session tool calls, errors, and token ledgers into a queryable global accumulation
- Inject a byte-deterministic low-cost summary into each turn's system prompt to stabilize prefix caching and cut token spend
Best For
- Developers running DSH intensively who value token efficiency
- Automated workflows that need persistent environment context across sessions
- Terminal users who want faster session starts with less redundant probing
Related Plugins
Traceable, searchable cross-session memory for AI agents that turns conversation knowledge into a typed knowledge graph and recalls relevant subgraphs instead of replaying full history, natively integrated with DeepSeek Harness.
Gives DSH AI cross-session long-term memory, to-do and skill management, plus multi-session orchestration, external AI delegation, and an infinite canvas that grows with you.
dsh-mnemon is a three-tier, pluggable, Agent-driven memory system for DeepSeek Harness, combining Runtime memory, Project Documents, and replaceable Memory Spaces with nine long-term providers.
An opinionated, zero-infrastructure file-based memory protocol: human-readable markdown files plus a bounded always-loaded index, curation discipline, and no database or embeddings.