AI Agent Hub
Back to plugins
🧠

dsh-missher-memory

Memory Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install Missher12/dsh-missher-memory

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install Missher12/dsh-missher-memory in the DeepSeek Harness terminal to install the plugin; source code is available at https://github.com/Missher12/dsh-missher-memory .

About this plugin

Across multi-day development sessions, architectural rationale, rejected approaches, progress, and next steps tend to scatter across chat history and vanish the moment a terminal closes. dsh-missher-memory adds a standalone, installable Super Memory layer to DeepSeek Harness that lets you retrieve decisions, progress snapshots, failed attempts, and follow-ups within a confirmed project binding, without modifying the Harness core or duplicating any existing legacy database.

Key capabilities: indexed semantic recall with automatic seven-day duplicate consolidation (sources are archived, never deleted, and a capsule rollback restores every source and FTS row); strict isolation between project memory and personal preferences; candidate capture enabled by default but only searchable after human review; automatic recall restricted to approved content that fails open on timeout or error; built-in credential and sensitive-data detection that discards the entire session buffer on a match.

Designed for developers running DeepSeek Harness 0.1.x / Desktop 0.3.8+ who work on multi-day, multi-worktree projects and need a local, searchable, reversible, and auditable memory structure without depending on an external vector store.

Use Cases

  • Recalling architectural decisions, progress, and failed paths across multi-day sessions
  • Isolating project memory and personal preferences across multiple worktrees
  • Turning scattered chat context into a local, searchable, and reversible memory structure

Best For

  • Individual developers running long-term projects on DeepSeek Harness
  • Engineering teams that need worktree-level context isolation and prefer local-first data
  • Technical buyers who want a pure-JavaScript plugin with no native addons or external vector store dependency