Language Policy / 语言策略: Detect language from input. Output primarily in that language, with these exceptions:
- Technical terms stay in original language — ZH user learns React: output uses "State", "Props", "Hooks" (not forced translations)
- Product/UI names stay in original — "点击 Settings 面板", "打开 File 菜单"
- Code, APIs, function names always in original —
useState(), useEffect()
- Topic names stay as user provided — "帮我学 React Hooks" keeps "React Hooks" in ZH output
- Explanatory prose matches user's language — sentences, paragraphs, analogies all in one language
- Never mix for no reason — don't insert random English words into Chinese sentences (e.g. NOT "这个概念需要 walkthrough 一下")
Detection: Pure EN→EN, pure ZH→ZH, mixed→dominant request language, code-heavy (50%+)→EN with ZH notes if user uses ZH, unclear→check path.json language field, other language→default EN but respect user's terms.
Core: Feynman Four Steps
1. Study → Push today's content 2. Explain → User explains in own words
- Find Gaps → Identify weak points 4. Simplify → Plain language/analogies, 3-5 sentences
Not one-way delivery — conversational learning. User "speaks it out", exposes blind spots.
Mode A: Multi-Day Learning | Mode B: Quick Learn
Mode A triggers: "help me learn xxx", "quick start xxx", "I want to learn xxx", "帮我学 xxx", "快速入门 xxx"
Mode B triggers: "explain this article", "read this book", "summarize this link", "帮我解读这篇文章", "快速读这本书"
Mode A: Systematic Learning Flow
User wants X → Search sources → Estimate days → Generate path → Create cron → Daily Feynman → Self-test → Advanced recs
Step 1: Search Sources
Search 5 dimensions, verify authority/recency, save to learning-data/{slug}/sources.md:
| Query | Purpose |
|---|
{topic} tutorial beginner 2024..2026 | Latest beginner guide |
{topic} documentation official | Official docs (⭐⭐⭐) |
{topic} best practices | Authoritative guide (⭐⭐) |
awesome {topic} / {topic} roadmap | Community curated (⭐⭐) |
{topic} video course | Video/audio (⭐⭐) |
Priority: ⭐⭐⭐ Official/GitHub Org → ⭐⭐ Known communities/courses → ⭐ Quality blogs → ❌ Reject unsigned/outdated/promotional.
Full strategy:
references/source-search.mdStep 2: Estimate Days & Generate Path
Formula: Base days = max(3, ceil(new concept count / 3))
| Modifier | Adjustment | Modifier | Adjustment |
|---|
| Zero foundation | ×1.3 | Related experience | ×0.7 |
| Hands-on needed | +2 days | Crash mode | max(3, -2) |
| Systematic | ×1.5 |
Assessment: Search concepts → Count → Apply → Confirm with user.
Generate path to learning-data/{slug}/path.json:
{
"topic": "xxx", "slug": "xxx", "total_days": 7, "created_at": "2026-04-06",
"status": "active", "current_day": 1, "daily_time_min": 30,
"preferred_time": "09:00", "timezone": "auto-detect", "push_channel": "webchat",
"learning_method": "feynman", "language": "en",
"days": [{"day": 1, "title": "Build Awareness", "keywords": ["overview", "introduction"],
"feynman_prompt": "Try to explain in your own words: What is {topic}? Why is it needed?",
"simplify_target": "Imagine explaining to someone who knows nothing about it.",
"completed": false, "completed_at": null}]
}
Helper script: python3 skills/quick-learn/scripts/learner.py path "{topic}" {days} --output learning-data/{slug}/path.json
Domain-specific patterns: references/learning-patterns.md
Step 3: Create Daily Cron
One cron per plan, plans don't interfere. Fixed framework at creation, real-time content at push.
{
"name": "quick-learn: {slug}",
"schedule": { "kind": "cron", "expr": "0 {hour} ", "tz": "auto-detect" },
"payload": { "kind": "agentTurn",
"message": "Check learning-data/{slug}/path.json, if status=active and current_day<=total_days:\n1. Read today's title\n2. Real-time search latest sources\n3. Organize structured notes (Step 4)\n4. Push to user\n5. Attach Feynman prompt\n6. current_day+=1\nIf completed, push completion message and disable this cron." },
"sessionTarget": "isolated", "enabled": true
}
Step 4: Daily Push Format
Don't just drop links! Fetch, read, then write structured summary. 500-800 words (excl. diagrams).
📚 Learning Plan | {topic} · Day {n}/{total}
🎯 Today's Topic: {title}
⏱ Estimated Time: {total_min} minutes
━━━━━━━━━━━━━━━━━━━━━━━━━━
I. Core Concepts (2-4 accessible paragraphs)
II. Key Points (with → Source)
III. Concept Relationships (one sentence or diagram)
IV. Diagram (Mermaid / ASCII fallback)
V. Real Examples (1-2 life analogies)
━━━━━━━━━━━━━━━━━━━━━━━━━━
📖 Further Reading + 🎧 Good for commute
━━━━━━━━━━━━━━━━━━━━━━━━━━
✏️ Now try in your own words: {feynman_prompt}
💡 Estimated {total_min} min. Short on time / want more, just say 👇
6 rules: Fetch before write, plain language with examples, cite sources, verify authority, verify recency (<2 yrs), control length.
Step 5: Skip & Difficulty
- "Skip Feynman" / 「跳过费曼」→ Respect, gently suggest trying next time
- "Didn't get it" / 「没太看懂」→ Lower difficulty, simpler analogies
Step 6: Mid-Plan Management
Abandon / 放弃: Empathize → Diagnose (too hard/easy/busy) → Offer solutions → If persists: summary-abandoned.md, status: abandoned, disable cron.
Restart / 重新开始: Go back day → modify current_day. Full restart → archive path.json as path-v1-backup.json, create new. Different angle → update sources.md.
Daily Adjustment / 每日调整:
| Scenario | Response |
|---|
| Only 15 min / 只有 15 分钟 | Core concepts + diagram, Feynman→1 question |
| Want more / 多学点 | Add optional + deeper, can skip 1 day |
| 3 days time_ratio<0.5 | Suggest: extend days / halve amount / pause |
Record actual duration in
path.json days array:
{"day": 1, "title": "...", "planned_time_min": 30, "actual_time_min": 15,
"time_ratio": 0.5, "adjusted_content": "core_only",
"completed": true, "completed_at": "2026-04-06T09:15:00"}
Step 7: Daily Log
Auto-record to learning-data/{slug}/daily-log.md and daily-log.json:
Completed At | Study Duration | Feynman Restate | Weak Points | Simplify Quality (1-5⭐) | Mood (Positive/Neutral/Exhausted/Want to quit)
Use: Reports, pace adjustment (consecutive "exhausted"→slow down), abandonment summary.
Step 8: Forgetting Curve Review
After each concept completion, push review cards at 1d / 3d / 7d intervals. Cron pushes card only; main session handles correct/wrong judgment and chain scheduling.
→ Format, cron config, state management: references/review.md
Step 9: Knowledge Map & Weekly Report
Knowledge Map / 概念地图: Every 2-3 days, generate Mermaid map showing mastered/learning/pending. Formula: progress% = mastered / total × 100.
Weekly Report / 学习周报: Auto weekly (Sun 20:00). Includes study time bar chart, completion progress, mood trend, Feynman quality stars. Zero-data week: show last activity + resume prompt.
→ Full formats, cron config: references/visualization.md
Step 10: Learning Style
First 2 days balanced, from day 3 auto-adjust based on behavior (diagram/text preference, Feynman quality, interaction depth). Manual anytime: "more diagrams/code/concise".
→ Full rules: references/feynman.md
Step 11: Advanced Recommendations
After self-test/plan completion: auto-generate 🥇recommended + 🥈alternative + 🥉expand suggestions based on topic dependencies and performance.
→ Logic and format: references/learning-patterns.md
Step 12: Progress Management
Single plan / 单计划:
| Says (EN) | Says (ZH) | Action |
|---|
| "Where am I?" | 「学到哪了」 | Read path.json + daily-log.md report |
| "Pause learning" | 「暂停学习」 | Disable cron quick-learn:{slug} |
| "Resume" | 「继续」 | Re-enable cron |
| "Move to 8pm" | 「调到晚上8点」 | Update cron schedule |
| "Abandon" | 「放弃」 | Summary → delete cron → mark abandoned |
| "Start over" | 「重新开始」 | Archive old, create new path.json |
Multi-plan overview / 多计划总览: Read all
learning-data//path.json, summarize topic/day/total/status.
📋 Active Learning Plans
- 🚀 React Hooks — Day 3/7 | 43% complete. Next: Core Architecture · Today 09:00
Concepts: 3 mastered, 2 learning, 4 pending
- 🚀 Docker — Day 1/5 | 20% complete. Next: Build Awareness · Pause scheduled
Concepts: 1 learning, 5 pending
Mode B: Quick Learn — Article/Book/Link
Quick learn data under learning-data/quick-{slug} (e.g. quick-article-abc123/, quick-book-xxx/).
learning-data/
├── react-intro/ ← Systematic
├── quick-article-abc123/ ← Quick (article)
└── quick-book-xxx/ ← Quick (book)
Output Template
📖 Quick Learn | {title}
📝 Type: {Article / Book / Blog / Docs} ⏱ ~{x} min
━━━━━━━━━━━━━━━━━━━━━━━━━━
{core idea in 1 sentence}
- Point + brief explanation
{Data, examples, cases}
- Actionable insight 1, 2
{Mermaid or ASCII}
{What's missing? Bias? Alternatives?}
━━━━━━━━━━━━━━━━━━━━━━━━━━
✏️ Feynman: {question based on content}
💡 Reply with understanding, or "done" to skip.
Save to learning-data/quick-{slug}/breakdown.md. Feynman follow-up same as Mode A (gap check → simplify).
Completion Rules / 完成规则
Mark completed: true when:
- User finishes Step 4 simplification AND rated ≥3⭐, OR
- User explicitly says "done"/"skip"/"move on", OR
- User abandons the plan
Do NOT mark complete after gap check (Step 4-2) alone — user must restate or explicitly skip.
Error Handling / 错误处理
Source URL inaccessible: try alternative → if all fail for a dimension, skip → record in sources.md → notify user content compiled from remaining verified sources.
by