dsh-project-memory

One of the biggest pain points when using AI coding assistants is their inability to remember project-specific context, conventions, or decisions across sessions. You often find yourself re-explaining 'we use pnpm for builds' or 'the API uses OAuth2' in every new chat, which is inefficient. The dsh-project-memory plugin solves this by providing true project-level persistent memory for deepseek-harness.

The plugin's core capability is to persist memory as plain Markdown files within your project's memory/ directory (with MEMORY.md as an index and *.md files for details). This makes all memory fully auditable, git-trackable, and diff-friendly, completely decoupled from the model's internal context. It offers a complete suite of tools for memory lifecycle management: intelligently loading relevant memory at session start (memory_upsert), storing user-directed decisions (memory_finalize), consolidating key conclusions at task end (memory_finalize), and securely forgetting outdated information (memory_forget). A built-in sensitive information interceptor also blocks storage of keys, passwords, and other secrets.

This plugin is ideal if you are a developer using deepseek-harness and want your AI assistant to truly understand and remember your project's evolving conventions, architectural decisions, and team preferences. It's particularly well-suited for projects that value traceability, collaboration, and security, as all memory changes are recorded as plain-text files in your version control history.