dsh-workspace-memory: A DSH plugin for project-scoped memory handoff in DeepSeek Harness
libraqg/dsh-workspace-memory
Project-scoped memory handoff for DeepSeek Harness: inject a curated brief from earlier conversations into a new session with one confirmation and exactly one model call.
dsh-workspace-memory is a DSH plugin for DeepSeek Harness that curates earlier conversations in the same project into an injectable handoff brief. It addresses the case where a fresh session's agent knows nothing about the previous conversation's goals, decisions, and pending work, so you stop re-explaining them. Its core abilities include a single model call after explicit confirmation, a pre-send view of source conversations and estimated token cost, project isolation via working-directory root resolution, reuse of existing compaction summaries, verbatim pinned notes, and auditable, resettable local storage.
How to Install
dsh plugin --profile web add @libraqg/dsh-workspace-memory- Category
- Data & Knowledge
- Platform
- DSH Plugin
- Author
- libraqg
- Distribution
- Plugin
dsh-workspace-memory Key Features
dsh-workspace-memory Repo Summary
What Does It Do?
Workspace Memory is a DSH plugin for DeepSeek Harness that turns earlier conversations in the same project into a curated handoff brief you can inject into a fresh session. It solves the problem of a new agent not knowing what the previous conversation did, so you no longer re-explain goals, decisions, files, and pending work. The plugin calls a model only once, after your explicit confirmation, and never makes silent background calls.
Core Features
- One confirmed model call: the only model-call path is the confirmed "Inject memory" action; page load, panel open, session creation, and turn end never trigger a call.
- Spend before you send: the confirmation dialog shows source conversations, the curator model, estimated input tokens, the output limit, and estimated injection tokens, and states that the project summary goes to the model provider.
- Project-scoped isolation: the project root is resolved from the session working directory via the registered DSH workspace root, nearest Git root, then the directory itself, so subdirectory sessions share a project while different projects stay separate.
- Compaction-aware reuse: when an earlier conversation already has a DSH compaction summary, the plugin reuses it instead of re-reading the raw transcript.
- Pinned notes stay verbatim: pinned notes.md content is inserted as-is, and the curator model is told not to rewrite or replace it.
- Auditable and resettable: brief, notes, and state files live in a local directory, and the sidebar panel shows the brief, sources, last model usage, and storage path, with a reset option per project.
How to Use This Plugin?
Open a new conversation in the same project inside DSH, then start "Inject memory" from the sidebar panel. The confirmation dialog lists the source conversations and estimated token cost; after you confirm, the plugin runs a single curation call, writes the handoff brief into the first message, and the normal request proceeds. If the curator route fails, times out, or returns an invalid brief, the plugin reports the error and injects nothing, without silent retries or model switching. The UI follows the DSH locale by default, and the confirmation layer offers a persisted Chinese/English toggle.
How to Troubleshoot This Plugin?
When the curation call fails, times out, or returns an invalid brief, the plugin reports the error and injects nothing, and it does not retry or switch models automatically; check the curator route configuration as described in the official docs, then confirm the injection again.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: libraqg/dsh-workspace-memory. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.
DSH Plugin FAQ
