ai-learning-accelerator

name	ai-learning-accelerator
description	AI-powered accelerated learning workflow that turns any subject into a structured, multi-modal learning sprint. Orchestrates context injection, multi-modal pre-warming, curriculum generation, 4-layer daily content packs (read/watch/practice/correct), AI sparring agent SOP design, and cross-validation (optionally with secondary model/tooling). Integrates with audio overview, image generation, and web search tools when available. Use when: user says 'learn X', 'help me study', 'learning plan', 'accelerated learning', 'how to learn', 'study sprint', 'from zero to X', '学习', '速学', '入门', '从零开始学', '学习计划', '加速学习', '怎么学', '帮我规划学习', '学习路线', 'AI辅助学习', '用AI学'.
metadata	{"version":"1.0","source":"local"}

AI Learning Accelerator

Orchestrate a complete AI-powered learning sprint for any subject.

Quick Start

1. Tell your AI assistant: "I want to learn [topic]"
2. The skill collects your background, time budget, and goals (Step 1)
3. It builds a trusted source pack and concept overview (Step 2)
4. It generates an N-day curriculum (Step 3)
5. It creates your first day's 4-layer learning pack (Step 4)

For quick-start mode, the curriculum spans all planned days but only Day 1's pack is generated immediately. Run the skill again to generate subsequent days.

See references/quick-start-example.md for a complete worked example.

Modes

Mode	Trigger	Deliverables
selection	User undecided on subject	learning_brief.md only
quick-start	Subject named, start now	learning_brief.md + source_pack.md + overview.md + curriculum.md + day-01.md
sprint	48h–3 day intensive	All of quick-start + remaining day-NN.md packs + agent_sop.md
standard	7–14 day plan	All of sprint + validation_notes.md

Default to quick-start if user names a specific subject. Ask only if ambiguous.

Per-mode output contract: Each mode produces ONLY the files listed. source_pack.md and overview.md are required for all modes except selection. Do NOT generate daily packs beyond what the mode requires.

Canonical Metadata Model

All files share these definitions. Do NOT deviate.

Depth levels (exactly 4, used everywhere):

• Survey — breadth over depth, build mental map
• Working — functional competence, can apply with reference
• Advanced — handles edge cases, can teach others
• Mastery — expert-level, can innovate and identify errors

Knowledge type tags (exactly 7, used in curriculum, daily packs, and SOP):

• factual | conceptual | procedural | symbolic | pattern | judgment | creative

Each curriculum item gets 1 primary tag and 1 secondary tag. No tertiary.

Confidence levels (for validation):

• C1-Verified — confirmed against primary source (T1/T2), citation provided
• C2-Consistent — multiple credible secondary sources agree, but no primary source confirmation yet
• C3-Unverified — cannot confirm from available sources, needs human review

Global Guardrails (Three Iron Rules)

Rule 1 — Exact-Symbol Rendering (enforced in Steps 4, 5): Use code blocks, SVG, HTML, or LaTeX for notation, formulas, diagrams, and structured symbols. NEVER use image generation for content requiring symbol fidelity. Image generation is ONLY for conceptual overviews and illustrations where visual style matters more than precision.

Rule 2 — Verifier Threshold (enforced in Steps 1, 4, 6): Each learning day includes a "verifier checkpoint" — what the learner should now catch if AI is wrong. The learning goal is becoming a reliable verifier, not just a knowledge holder.

Rule 3 — Context Engineering (enforced in Steps 1, 3, 5): Every major output is generated from an explicit template. No freestyle prompting. All templates are in references/.

Workflow (6 Steps)

Step 1: Intake + Scope Decision

Collect constraints using the template in references/intake-and-scope.md.

For selection mode: Subject field is "undecided" — run the decision flow in that reference file, then output learning_brief.md and stop.

For all other modes: All 6 fields required. Output learning_brief.md with scope boundaries, depth calibration (using canonical depth levels), and success milestones.

→ Rule 2: Include "known AI weaknesses in this domain" in the brief. → Rule 3: Use intake template from references/intake-and-scope.md verbatim.

learning_brief.md structure:

The brief is assembled from two sections, both defined in references/intake-and-scope.md:

1. Intake data — the 6-field intake template (Subject, Background, Scenario, Time Budget, Success Criteria, Preferred Modalities)
2. Focus Decision — the exact ## Focus Decision block from that reference (Scope Boundaries, Depth Calibration, Success Milestones, Known AI Weaknesses, Daily Verifier Seeds)

Do NOT define a separate schema here. The reference file is the single source of truth for learning_brief.md structure.

Step 2: Source Pack + Pre-Warming

Build a trusted source pack. See source trust ladder below.

Source Trust Ladder:

Level	Type	Rule
T1	Official docs, textbooks, peer-reviewed	Primary reference, cite directly
T2	Established courses, expert content	Cross-ref with T1
T3	Community content, blog posts	Verify claims before using
T4	AI-generated content	Always cross-validate, never sole source

source_pack.md schema:

# Source Pack
## T1 Sources (Primary)
- [Source name] — [URL/reference] — [covers what topics]
## T2 Sources (Secondary)
- [Source name] — [URL/reference] — [covers what topics]
## T3 Sources (Community) — use with caution
- [Source name] — [URL/reference] — [why included]
## Contested Topics
- [Topic] — [school A says X, school B says Y] — [recommendation]

Pre-warming output — generate overview.md:

Tool Available?	Action
Audio overview tool (e.g., NotebookLM)	Generate audio overview from source materials
Image generation tool	Generate conceptual overview image (NOT symbolic — Rule 1)
No tools available	Text-only concept map + glossary of key terms

overview.md schema:

# Overview: [Subject]
## Concept Map (text)
[Central concept] → [branch 1] → [sub-concepts]
                  → [branch 2] → [sub-concepts]
## Key Terms Glossary
- **[Term]**: [Definition]
## 3 Things AI Often Gets Wrong About This Subject
1. [Specific error pattern]
2. [Specific error pattern]
3. [Specific error pattern]
## Audio Overview
[Link to audio overview if generated, or "Not generated — text overview above"]

Step 3: Generate N-Day Curriculum

Generate curriculum.md from template:

# Curriculum: [Subject] — [N] Days

## Day [N]: [Topic Title]
- Objectives: [2-3, observable/testable]
- Primary tag: [one of 7 canonical types]
- Secondary tag: [one of 7 canonical types]
- Time estimate: [hours]
- Prerequisites: [Day X or "none"]
- Verifier checkpoint: [what learner can now catch]

Adaptation rules (applied during generation):

• Time tight → core concepts only, mark "optional" topics
• Background ≥ 5 → skip fundamentals, start at Working depth
• Practical output goal → weight procedural + pattern tags

→ Rule 3: Generate from this template, not freestyle.

Step 4: Generate Daily 4-Layer Packs

For each day, produce day-NN.md following the contract in references/daily-pack-contract.md.

4 layers: Read, Watch, Practice, Correct. Exercise types matched via references/knowledge-to-exercise-map.md.

→ Rule 1: All formulas/notation in code blocks. → Rule 2: Each day includes 1 "verifier challenge" (deliberately tricky AI-error-prone item).

Step 5: Generate Sparring Agent SOP

Mode gate: Only for sprint and standard modes. Skip for selection and quick-start.

Produce agent_sop.md using the template in references/agent-sop-template.md.

The SOP uses primary and secondary tags from the curriculum (NOT tertiary). Quiz weighting: primary 50%, secondary 30%, verifier challenges 20%.

→ Rule 3: SOP itself follows context-engineering — explicit steps, forced output structure.

Step 6: Cross-Validate + Recalibrate

Mode gate: Only for standard mode. Skip for all other modes.

Follow the protocol in references/validation-protocol.md. Produce validation_notes.md.

Tool Integration

Step	Tool	What It Does	If Unavailable
2	Audio overview tool (e.g., NotebookLM)	Audio overview from sources	Text concept map + glossary
4	Slide/image generation tool	PPT slide images from content outline	Text concept map + curated links
4	Image generation tool	Conceptual illustrations (NOT symbolic)	ASCII/text diagrams
6	LLM with web search (e.g., Grok, Perplexity)	Fact-checking via web search	Web search tool or direct check against T1/T2 sources; otherwise mark C3-Unverified
6	Secondary LLM	Code/formula cross-verification	Direct check against T1/T2 sources

All tools are optional accelerators, not blockers. If a tool is unavailable, use the fallback. Never halt the workflow because a tool is missing.

When NOT to Use

• Quick factual question (just answer it)
• Looking for a document or resource (use web search)
• Automating a repetitive task (use a task automation tool)
• Medical/legal advice (consult domain-specific tools or a professional)