AI Agent Hub
Back to plugins
runtime36 preview

runtime36

Memory Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install 398894496-arch/runtime36

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

Install the runtime36 plugin in DeepSeek Harness using the full source URL https://github.com/398894496-arch/runtime36.

About this plugin

runtime36 addresses a common problem for coding agents: project decisions, corrections, and lessons learned get scattered across chat logs, so the next session starts by re-explaining what was already understood. It uses an Obsidian vault as a durable knowledge base, letting agents seal the day, distill useful material, and promote reliable methods instead of dumping raw chat history into the next prompt.

Its core is a knowledge workflow for agents: daily logs and distilled summaries are kept separate, promotion requires explicit gates, and retrieval relies on the user's own short nouns and alias table to point to a canonical page with deterministic verification. It behaves like a writing and routing system, where you maintain your nouns, projects, and corrections while agents find stable method pages and distinguish formal, provisional, and pending-approval knowledge.

It is best for developers and agent users who already keep an Obsidian vault, or want to build one, especially people working across Cursor, Codex, Claude Code, and DeepSeek Harness who need durable memory for recurring decisions. If the goal is automatic semantic search, hosted memory, or chat content injected directly into the next prompt, this plugin is not the right fit; it is for people who want memory to be locatable, maintainable, and auditable over time.

Screenshots

Use Cases

  • Seal and distill agent chats into reusable methods each day.
  • Locate experience pages precisely with short nouns and an alias table.
  • Update canonical pages after corrections so later agents read the current version.

Best For

  • Developers who maintain an Obsidian vault and want durable memory for agents.
  • Agent users sharing knowledge routes across Cursor, Codex, Claude Code, or DeepSeek Harness.
  • People willing to maintain noun tables and turn repeated decisions into methods.