yuezengwu/dsh-explain
DSH local-first learning mode plug-in: cross-session global learning thread, explanation by source, ExplainContext, compression and diagnostic settings interface
dsh-explain is a learning-mode plugin that turns useful concepts from completed work sessions into structured explanations, keeps them in one global learning thread, and adapts future explanations to what the user already knows. Entry points: /explain <request> (current session as bounded source context), Explain selected text (editable /explain --selection draft, never auto-submits), Learn from this answer (editable draft tied to the exact finalized assistant turn), and automatic evaluation after eligible completed turns within a configured budget. Each learning card answers three questions — What is it? Why does it matter? What is the common pitfall? — with Got it / Not yet feedback and rephrasing even if the source session is later deleted. /explain on|off|status control the runtime from the composer. It observes only future completed top-level turns; it does not scan existing history.
Install
npx @deepseek-ai/dsh@0.1.0-rc.8 plugin --profile web add github:yuezengwu/dsh-explainExplain currently targets DSH 0.1.0-rc.8 — npx @deepseek-ai/dsh@0.1.0-rc.8 plugin --profile web add github:yuezengwu/dsh-explain, then npx @deepseek-ai/dsh@0.1.0-rc.8 web. Open Settings → Learning, select an auxiliary provider and model, enable learning mode, and save. Git-hosted plugins build during installation; if pnpm requests build approval, add the printed dsh-explain entry to the profile's pnpm-workspace.yaml and repeat the install command.
Compatibility
DeepSeek Harness 0.1.0-rc.8. The primary agent stays untouched: Explain uses its own model calls, scheduler, context, and local SQLite database. Local-first (data-local-first badge). Every $DSH_HOME owns exactly one global learning thread; resumes and forks never copy learning state. Default autonomous budget is 50 requests per rolling 24 hours and survives restarts.
Details
- Repo: yuezengwu/dsh-explain
- Category: Agent Capabilities
- Stars: 11
- Version: Git-source plugin; MIT; DSH 0.1.0-rc.8 (badge)
- Last push: 2026-09-18
- First seen: 2026-08-12
Recent updates
The current README documents the learning-loop concept, the four entry points, the three-question learning card, the single-global-thread model, the autonomous budget, and the Quick start for DSH 0.1.0-rc.8.
FAQ
- Does it scan my existing history?
- No — Explain observes only future completed top-level turns; it does not scan existing history.
- Does it touch the main agent?
- No — the primary agent stays untouched; Explain uses its own model calls, scheduler, context, and local SQLite database.
- Do forks copy learning state?
- No — every $DSH_HOME owns exactly one learning thread; resumes and forks never copy learning state, and each source session has at most one explanation awaiting feedback.
Alternatives
csyangwen/dsh-memory-evolve · ZK-Andy/dsh-continual-evolve · ZSeven-W/dsh-noema