Learning Mode agent preset: a coding agent that teaches while coding — concrete scenario-grounded explanations, Socratic guidance, and TODO(你) practice blanks, modeled on Claude Code's Learning output style; installable via dsh plugin add (dsh-learning-mode on npm).
Install
# from npm (prebuilt)
dsh plugin --profile web add dsh-learning-mode
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:CHplus0/dsh-learning-mode#path:/plugin
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time. Only install sources you trust, and pin a commit (github:owner/repo#sha).
README
English | 中文
An agent preset for DeepSeek Harness (DSH) that teaches while coding, modeled on Claude Code's official Learning output style: explain concretely with usage scenarios, guide your thinking with questions, and leave explicit practice blanks for you to do by hand.
Learning output style (Claude Code): "Collaborative, learn-by-doing mode where Claude will not only share 'Insights' while coding, but also ask you to contribute small, strategic pieces of code yourself."
Three Pillars
| Pillar | Behavior |
|---|---|
| A · Concrete, scenario-grounded explanations | Flexible, task-tied explanations: everyday analogy (boundaries marked when applicable) + scenario grounding (when/which/why results differ), used as needed; simple concepts get a sentence or two, complex ones get expanded. Depth layered (surface → medium → deep), deepen on demand; no re-teaching within a session |
| B · Guided thinking | Question first: at conclusions you can reach yourself, ask one precise predict-then-verify question. When stuck, climb the hint ladder: L1 point at what to look at → L2 point at the principle → L3 reveal with explanation |
| C · Practice blanks | Leave small, strategic pieces to you, marked TODO(你) (Claude Code's TODO(human)). Small, strategic, tied to what you're actually doing, self-verifiable; never blank safety-critical, irreversible, or correctness-critical steps |
Interaction protocol: Teaching-first by default; "just do it / no time / asap" switches to direct mode; asks your familiarity level (beginner/intermediate/advanced) once at the start; ≥2 failed attempts downgrades to a guided reveal. Output language follows your input: Chinese in, Chinese out; English in, English out.
Examples
Teaching DeepSeek Harness's principles — the verbatim output of a real learning-mode session — a full teaching turn showing how concrete explanations, guided thinking, a context-tied TODO(你), and the one-time opening calibration work together. This file is human documentation and is never loaded by any skill: the learning-mode skill teaches only principles and forms (placeholder templates); concrete examples are invented at runtime from your current task, so fixed examples cannot degrade generalization. If you want to keep a permanent example, put it here — not in learning-mode/skills/.
Install
Requires DSH 0.1.0-rc.x (a deployment with profiles/ under ~/.dsh).
# Option 1: clone and copy
git clone https://github.com/CHplus0/dsh-learning-mode.git
cp -r dsh-learning-mode/learning-mode ~/.dsh/.agent-presets/
# Option 2: run the installer
bash dsh-learning-mode/install.sh
# Option 3: install the npm bundle (auto-installs the preset)
dsh plugin --profile web add dsh-learning-mode
Then open the DSH web UI, start a new session and pick 学习模式 (Learning Mode) — no restart needed.
Customization
- Tone & identity: edit
learning-mode/agent.cordis.yml→persona.text. - Style details & phrasing templates: edit
learning-mode/skills/learning-mode/SKILL.md. - Rename: edit only
nameinlearning-mode/preset.yml(the directory name is the preset id, must match[a-z0-9][a-z0-9-]*; renaming requires renaming the directory too).
How it works
agent.cordis.ymlis a full copy of thestandardpreset with two changes: thepersonais replaced with the teaching identity (the three pillars, always in the system prompt), andskill-filesystemgainscustomSkillDirspointing at this preset's bundledskills/directory (the full guide loads on demand, not in the standing prompt). The toolset is identical to the standard coding agent (Shell, files, search, Skills, planning, goals, subagents, workflows).
License
MIT © 2026 CHplus0. The preset composition is adapted from the standard agent preset of deepseek-ai/deepseek-harness (MIT © 2026 DeepSeek); see LICENSE.
Links
More in this category
superdesigndev/treg★ 460
Tool catalog for agents: search ~2,600 external endpoints (SEO and SERP, backlinks, social, people and company enrichment, ad libraries, scraping) by the task you want done, read each one's parameters and per-call price, then call it with the credential injected server-side. Ships the skill plus an MCP row that stays disabled until TREG_TOKEN is set.
zhaoolee/notes#dsh-plugin★ 147
Export DSH conversations as Smartisan Notes-style PNGs, or create and update Markdown notes in a configured account-scoped workspace.
taxueseek/argo#dsh-plugin★ 99
Search built for agents: multilingual coverage across web, academic, code, shopping, finance, news, and encyclopedias.
omdsh-dev/dsh-data-agent★ 52
Let the AI connect to databases and write SQL for you.
heartleo/hn-cli#hacker-news★ 50
Hacker News tools for feeds, discussion threads, search, and user profiles.
geml-spec/geml#integrations/dsh-plugin★ 24
Block-addressed document editing: an MCP server exposing geml_get / geml_set / geml_check and friends, so an agent reads or rewrites one addressed block of a Markdown or GEML document instead of the whole file. Ships the GEML authoring skill and a code-graph skill that builds and navigates a project call graph as GEML codemaps.