dsh-hermes-bridge
Run the following command in DeepSeek Harness:
dsh plugin install aalvsz/dsh-hermes-bridge
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install aalvsz/dsh-hermes-bridge in DeepSeek Harness to install the plugin; source code is available at https://github.com/aalvsz/dsh-hermes-bridge .
About this plugin
DSH already covers the agent loop, tool calling, subagents, and model routing, but it lacks the adaptive layer that lets an agent keep improving across sessions. dsh-hermes-bridge rewrites that layer natively in JavaScript, eliminating the Python runtime and external Hermes installation so that memory, skills, and trajectory tooling live entirely inside the DSH plugin system.
The core capabilities include persistent memory backed by MEMORY.md and USER.md (frozen into a system-prompt snapshot), a reusable skill catalog with list, view, and manage actions, skill authoring via /hermes-learn, optional background-review and curator subagents, and a full RL trajectory pipeline that converts each completed conversation into ShareGPT format, appends it to a JSONL file, and compresses it within a token budget by preserving head and tail turns while summarizing the middle — producing SFT/DPO-ready fine-tuning data out of the box.
It is well suited for developers working in DSH who need cross-session memory, skill reuse, or RL data engineering. Every tool is namespaced under hermes_* to prevent collisions, background review, curator, and trajectory saving are off by default, and file permissions are tightened to 0600/0700, so the plugin is safe to enable immediately.
Use Cases
- Preserve user preferences and context across sessions without re-explanation
- Auto-capture and compress completed conversations into SFT/DPO-ready fine-tuning data
- Build a reusable skill catalog and continuously author new skills via /hermes-learn
Best For
- Developers who need cross-session memory and skill reuse in DSH
- ML engineers who require an RL trajectory collection and compression pipeline
- Ops or platform teams who prefer pure-JS deployment without introducing a Python runtime
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.