dsh-memory-evolve: Cross-Session Memory and Self-Evolution for DeepSeek Harness

What Does It Do?

It is a DSH plugin for DeepSeek Harness that turns session context into long-term working memory spanning multiple conversations. It solves the problem of an AI forgetting your preferences and earlier decisions whenever you switch projects, wait a few days, or open a new session. Its core capabilities cover five-track memory, git branch awareness, skill self-evolution, four-track todos, session orchestration, and dispatching work to external AI agents. Maintained by csyangwen under the MIT license, it requires zero core modifications and zero runtime dependencies, so it works as soon as it is installed and leaves nothing behind when removed.

Core Features

• Five-track memory: user profile, global facts, project key memory, project logs, and daily logs are stored in layers, with progress written automatically each turn and important conclusions saved only after your confirmation, optionally scoped to a single git branch.
• Memory sync: uses your code repository or a dedicated shared memory repository as the remote, so multiple machines recognize the same project automatically and keep project memory, logs, archives, and todos consistent, while projects without sync stay purely local.
• Four-track todos and local search: life, work, project, and daily todos are isolated by working directory with priority and due dates, and the plugin can also search local documents by filename or content.
• Session orchestration and collaboration rooms: a main session can spin up multiple sub-sessions into one collaboration room, member status changes trigger notifications, room messages support images, and dispatching or nudging work stays under your deliberate control.
• Skill self-evolution and prompt management: a built-in skill manager and prompt manager, combined with in-turn self-review, let skills and prompts accumulate as you keep using them.

How to Use This Plugin?

After enabling it in DSH, open any session and you will see capability tabs for memory, skills, todos, COI scheduling, session broadcast, prompts, and memory sync, while the AI gains tools for memory read/write, todos, skill management, and session orchestration. Day-to-day use is just normal conversation: progress is written to the project log and today's log at the end of each turn. When you spot an important fact, say "remember this: the deployment port for this project is 8080" and it enters a pending queue that takes effect only after you confirm. To recover context, ask "check memory, what architecture decision did we settle on last time" and it picks up seamlessly; once memory grows, you can ask the AI to archive an entry so it stops being injected into context. Most capabilities are off by default, so turn them on one by one in Memory Evolve settings, and enable the related ones together to form a complete pipeline.

How to Troubleshoot This Plugin?

If the plugin misbehaves and prevents DSH from starting, mark it as disabled in the profile configuration and undo that once it is fixed, without uninstalling. If the loader reports a duplicate loader entry id, it means the plugin was inserted manually following older documentation and now duplicates the bundle's automatic registration, so remove that manual insert line from the profile patch. Upgrading from an older version requires the same cleanup of manual insert records.