dsh-plugins DSH plugin

This repository provides three ecosystem plugins for DeepSeek Harness: bounded file-backed memory (dsh-memory), skill authoring (dsh-skill-create), and automatic skill usage tracking with curation (dsh-skill-curator). Together they form the full learning loop inspired by hermes-agent. The project is archived and not maintained, but the code remains for reference.

skills-workflows

How to install the dsh-plugins DSH plugin

dsh plugin --profile web add @hyls9527/dsh-memory

Copying does not run this command. Review the repository and version before installing the dsh-plugins DSH plugin.

dsh-plugins DSH plugin data source

dsh-plugins DSH plugin snapshot date: Aug 16, 2026

discovered

What the dsh-plugins DSH plugin can do

Bounded, file-backed memory with MEMORY.md/USER.md, supporting add/replace/remove/batch operations via a single memory tool
Skill creation tool that persists reusable procedures as user skills (create/update)
Automatic skill-tool usage tracking and a curator tool to report usage, archive unused skills, and restore them
Prefix-stable prompts: memory snapshot is injected at session start, mid-session writes never modify it, preserving provider KV caches
Atomic writes with drift protection and cross-process lock for memory file integrity

Where the dsh-plugins DSH plugin fits

Give the model persistent memory across sessions without relying on external databases
Enable the model to create and update its own reusable skills during a conversation
Automatically manage a growing skill library: track usage, archive stale skills, and restore when needed
Build a self-improving loop where the model remembers, creates, and curates its own tools
Experiment with hermes-agent mechanisms ported to DeepSeek Harness as a research reference

Who the dsh-plugins DSH plugin is for

DSH users who want cross-session memory and skill lifecycle features missing from the official harness
Researchers and developers studying how to extend DSH with external bundles

dsh-plugins DSH plugin limitations

Project is archived and not maintained; no further updates or bug fixes are planned
Experimental prototype: may have stability issues and is not recommended for production use
Only tested against DeepSeek Harness 0.1.0-rc; compatibility with newer versions is not guaranteed
Memory system uses file-based storage (MEMORY.md/USER.md) and is not suitable for high-frequency writes
Skill creation and curation rely on the model's ability to use the provided tools; results may vary

dsh-plugins DSH plugin: from the repository README

Quoted from the hyls9527/dsh-plugins README, the upstream source of the dsh-plugins DSH plugin. Copyright remains with the original authors.

> **ARCHIVED — experimental prototype, not maintained.** > > These plugins (memory, skill-create, skill-curator) were built as a research exercise porting > [hermes-agent](https://github.com/NousResearch/hermes-agent) mechanisms to DeepSeek Harness. They work, > but the exercise concluded they are not useful: DeepSeek Harness deliberately keeps session state in the > session log and treats skills as user-authored, vetted inputs; the official repo does not accept external > PRs and has repeatedly, in writing, chosen not to build these mechanisms. The code remains here for > reference. No further maintenance is planned. English | [中文](README.zh.md) Ecosystem plugins for [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness), porting two of the most important mechanisms from [NousResearch/hermes-agent](https://github.com/NousResearch/hermes-agent): the bounded cross-session memory system and the skill lifecycle curator. | Package | What it does | |---|---| | [`@hyls9527/dsh-memory`](packages/memory) | Bounded, file-backed `MEMORY.md` / `USER.md` memory: `§`-separated entries under character limits, a frozen snapshot injected at each session start, one `memory` tool (`a

Read the full READMERepository license: MIT

dsh-plugins DSH plugin questions

How do I install these plugins?

You can install each plugin via `dsh plugin --profile web add @hyls9527/dsh-memory` (and similarly for the other two). If you cloned the repo, you can also add from local paths: `dsh plugin --profile web add ./packages/memory`. The plugins are added to your profile's bundles list.

Are these plugins still maintained?

No. The repository is archived and explicitly marked as not maintained. The code is provided as a research reference. No further updates, bug fixes, or support will be provided.

What version of DeepSeek Harness do they work with?

They were verified against DeepSeek Harness 0.1.0-rc. Compatibility with later versions is not guaranteed. You may need to adjust the configuration or code if you use a newer DSH release.

Can I customize the memory character limit or nudge interval?

Yes. You can override plugin configuration in your profile's `cordis.patch.yml`. For example, set `memoryCharLimit: 4000` or `nudgeInterval: 0` for the memory plugin. See the examples/cordis.yml file in the repository for a complete overlay.

Why was this project archived?

The authors concluded that the mechanisms are not useful for DeepSeek Harness because DSH deliberately keeps session state in the session log and treats skills as user-authored, vetted inputs. The official DSH repo does not accept external PRs for these features. The exercise was a research prototype that ended.