dsh-state
Run the following command in DeepSeek Harness:
dsh plugin install xiyiyiru/dsh-state
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install xiyiyiru/dsh-state in the DeepSeek Harness terminal to install; source code at https://github.com/xiyiyiru/dsh-state
About this plugin
The quiet killer in long-running sessions is context compaction: a root cause just confirmed, a direction the user just signed off on, both vanish the moment the window shrinks. dsh-state takes a deliberately simple route. Working facts are written to plain-text files under .mycel/state/, so they survive compaction and resume; nothing is auto-injected into any request. If you want the notebook, you call read_state; if you do not, the cost is zero.
The core is five tools. add_state appends one note (corrections are new lines, there is no edit verb), read_state returns the full text and flags a compress hint past 8K, and compact_state rewrites the whole file as a summary. focus_task and focus_complete form a task stack: lock a multi-step task, solve blockers as sub-frames, and pop back to the main thread when done. All state is keyed by session id, so two conversations in the same workspace never collide.
It is aimed at dsh agent users who run long, multi-step tasks and regularly hit compaction boundaries. No extra services, no schema to migrate, no session events written. The state directory is the single source of truth, and that is the entire dependency surface.
Use Cases
- Multi-step task hits context compaction mid-flight and must recover confirmed working facts
- A long conversation where the user made key decisions or corrections that persist across steps
- A task gets interrupted; progress is recorded and the agent resumes from the last frame
- A blocker is pushed onto the stack, solved, and the agent pops back to the main task
Best For
- dsh agent developers running long, multi-step workflows
- Users who frequently lose working memory to context compaction
- Minimalists who prefer plain-file state with zero extra services
Related Plugins
Memory layer for coding agents that indexes local session history and auto-recalls relevant context before edits, commands, and failures, with no manual search needed.
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.