| name | deep-learning |
| description | All-in-One Deep Reading tool for Zettelkasten. Designed for deeply digesting books, long-form articles, research reports, and academic papers. Use when a user wants to deeply digest a book/long article/report/paper and build a knowledge network, producing structure notes and atomic notes. Blends Mortimer Adler (structure), Feynman (clarity), Luhmann (network), Pragmatist (actionability), and Critics (debate). Enforces High-Fidelity case preservation and Actionable tool extraction. Keywords: deep reading, structure note, deep learning, Zettelkasten. |
Deep Reading
Core Philosophy: Not just understanding the world (Understand), but changing it (Act).
Scope: Book, Long-form Article, Research Report, Academic Paper.
The Council (Expert Personas)
- Mortimer Adler (The Architect): Responsible for structure โ extracting core propositions and logic trees.
- The Pragmatist (The Engineer): Responsible for actionability โ extracting executable SOPs, templates, and checklists.
- Richard Feynman (The Teacher): Responsible for clarity โ ensuring concepts are demystified and explained in plain language.
- Niklas Luhmann (The Librarian): Responsible for connectivity โ ensuring knowledge is networked and grows organically.
- The Critics (The Stress Testers): Musk, Socrates, Munger โ responsible for stress-testing and debate.
The Iron Rules
- Always Deep: Regardless of input length, always process at the highest level (structure + tools + debate).
- Case Fidelity:
- Source text available: No paraphrasing allowed; atomic notes involving cases or studies must preserve specific numbers, authors/institutions, timelines, and direct quotes (page numbers recommended).
- No source text, only summaries or memory: Mark the note with
Source: book/summary, not verified against original text; preserve identifiable proper nouns and conclusions; fabricating details is prohibited; Feynman review marks "Case Fidelity: Partial (no source text)".
- No Vague Verbs: Verbs like "optimize," "strengthen," "appropriately" are banned. Every action must be converted into a specific action or quantifiable metric.
- Metadata Mandatory: All notes must include YAML Frontmatter (type, tags, links). No exceptions.
Storage Rules
-
Default Location: Daily/YYYY/MM/DD (or your vault's daily folder, e.g. 05_ๆฏๆฅ่ฎฐๅฝ/YYYY/MM/DD)
- Get the current date (YYYY, MM, DD); create the date folder if it doesn't exist; create a task folder for this session.
- Task folder naming: Default to
[Title]_structure_note.md.
- Structure note naming: Default to
YYYYMMDD_00_[Title]_structure_note.md.
-
Structure note onboarding (choose one, see Phase 6):
- Option A: If the target index has an
## Inbox section, append this book's entry under Inbox (e.g. - [[This Structure Note]] โ Book Title, YYYYMMDD).
- Option B: If the target index has no Inbox, create a new Inbox section.
-
Index note onboarding (see Phase 2.5): Mount the new index note to an existing index, and move it to your index directory (e.g. Index/ or 03_็ดขๅผ/); see Phase 2.5 for Inbox/entry format.
The Workflow
Execution order: Phase 0 โ 1 โ 2 โ 2.5 โ 3 โ 4 โ 5 โ 6 โ 6.5 (Workflow Audit, mandatory). No skipping; Phase 2.5 must execute immediately after Phase 2; Phase 6.5 must execute immediately after Phase 6.
Phase 0: Pre-game Plan
Produce an execution plan before starting Phase 1; save as YYYYMMDD_01_[BookTitle]_execution_plan.md (same directory as the structure note), and link this file under the Preparation section of task.md.
Must include:
- TODO List (โฅ6 items): e.g. full-book argument skeleton, key concepts, argument chains, framework extraction, method extraction, case verification, critical review, network linking.
- Context: Reading intent (What problem am I trying to solve?).
Phase 1: Overview & Structure
Agent: Mortimer Adler
- Task: Create the structure note; output must conform to
templates/structure_note_template.md, including core propositions and supporting logic chains.
- Note: Invoke
structure-note skill if available.
Phase 2: Index Design
Agent: Niklas Luhmann
"Don't ask what category it belongs to โ ask who it talks to."
- Task: Create the index note for this book; output must conform to
templates/index_note_template.md, including keywords and multiple entry points.
- Note: Invoke
index-note skill if available.
Phase 2.5: Index Note Onboarding
Agent: Niklas Luhmann
Execute immediately after index note creation; ensure the new index is accepted into the knowledge network and physically filed.
- Mount to existing index: In your index directory (e.g.
Index/), select one or more existing indexes related to this book's topic, and add an entry pointing to this new index note (e.g. under ## Inbox append - [[This Index Note]] โ Book Title/Topic, YYYYMMDD, or add to the index's "Theme Structure / Sub-index" section). Ensure bidirectional linking: the new index note's footer should also link back to the parent index.
- Move to index directory: Move the index note file from the current task directory (e.g.
Daily/...) to the appropriate location in your index directory:
- Use
file-organize skill if available: determine whether it goes at root level (alongside existing hubs) or in a topic subfolder.
- After moving,
[[This Index Note]] links in the parent index remain valid (filename-only, path-independent).
- (Multi-index mount check deferred to Phase 6): Whether notes mentioned in this index should also be mounted to other indexes โ check and add entries in Phase 6.
Phase 3: Recursive Growth
Agent: Luhmann & Feynman
Core innovation: Create while discovering (Luhmann Scan).
Process:
- Round 1 (Skeleton): Starting from the structure note, create all explicitly linked Atomic Notes (Concepts).
- Luhmann Scan (mandatory for every note): See
references/luhmann_scan.md. Three checks โ prerequisites, potential connections, methodology discovery (if a concept includes an executable "how" โ log in task.md, create a detailed Method Note in Phase 4). Record format in task.md: Prerequisites โ Round 2: [[A]]. Connections โ Round 3: [[B]]. Methods โ [[MethodName]] (Phase 4).
- Round 2 (Flesh): Create the newly discovered notes, continue Luhmann Scan.
- Round 3+ (Edges): Until complete or out of scope.
Atomic Note Specification:
- Template:
templates/atomic_note_template.md
- Focus: Definition, Mechanism, Context.
- Acceptance: Feynman Test (a layperson can understand it).
Phase 4: Methodology Consolidation
Agent: The Pragmatist
Turn methods discovered in Phase 3 into Method Notes.
Method Note Specification (high priority):
- Template:
templates/method_note_template.md
- Focus: Executable Steps (SOP), Templates, Checklists.
- Constraints:
- No Vague Verbs: Every step must be concrete.
- Mechanism & Leverage: Why does it work?
- MVE: Minimum Viable Experiment as the next step.
- Practitioner-grade requirements ("Operations Manual" not "Academic Exchange"):
- Steps in tables: If a method has multiple steps, each step uses a
| Step | What | How | Why | table clearly separating action items, specific operations, and purpose/mechanism; pure paragraph descriptions of steps are prohibited.
- Inline troubleshooting: Embed "Common Issue โ Immediate Fix" after each step, rather than collecting all pitfalls at the end; the final Pitfalls section serves as a global anti-pattern summary.
- Self-assessment checklist: Beyond the output acceptance checklist (did the result meet the bar?), add a "Did I execute correctly?" self-check on execution behavior (e.g. process quality, judgment quality, collaboration quality โ adjust dimensions by domain).
- Pre-flight section: Beyond prerequisites, include an independent "Pre-flight Checklist" โ things to confirm before starting (environment, materials, mindset, etc.).
Phase 5: Final Review
Agent: Richard Feynman
Review the entire knowledge network through the Feynman lens:
- De-jargon check: Have all technical terms been "translated" into everyday language?
- Metaphor check: Do complex concepts have appropriate metaphors?
- Logic check: Are there any breaks in the argument chain?
- Topology check: Are there any "surprise" connections?
Phase 6: Network Review
Agent: Niklas Luhmann
- Link check: Confirm every note (excluding parent-child structure notes) has at least 2 bidirectional links.
- Onboarding: Complete index onboarding in relevant indexes under your index directory โ create
## Inbox if none exists; file structure notes and atomic notes under appropriate keyword/topic indexes. Invoke index-note mode 3 (content onboarding) if available for single or batch onboarding.
- Multi-index mount check: For every note mentioned in the index (those linked in the keyword section and theme structure section), evaluate whether it should also be mounted to other indexes (different lookup intents); if a note is a quality entry point for another topic, add an entry in the corresponding index to enable multi-path discoverability. Output a brief list:
[[NoteA]] โ already in Index_X; recommend adding to Index_Y (reason).
Phase 6.5: Workflow Audit (Mandatory)
Must execute after Phase 6 is complete. Invoke the workflow-audit skill to perform a Deming + Gawande-perspective execution completeness check and systematic closure:
- Audit target: This skill (deep-learning); artifacts: all task directory outputs (execution plan, structure note, index, atomic notes, task.md, etc.).
- Output: Save an audit report (
YYYYMMDD_[TaskName]_workflow_audit_report.md), including item-by-item checklist, system closure, DoD sign-off, multi-index mount list; any item marked โ must be remediated until all DoD items pass.
- workflow-audit main file:
.cursor/skills/workflow-audit/SKILL.md; report template: workflow-audit/references/audit_report_template.md.
Cannot be skipped. The workflow is considered incomplete until the audit passes and all gaps are remediated.
Task Tracking
Maintain progress in task.md:
# Preparation
- [ ] Pre-game Plan Created (โฅ6 TODOs + Context; in-conversation block or saved file)
# Structure & Index
- [ ] Structure Note Created (Adler)
- [ ] Index Note Created (Luhmann)
- [ ] Index Note Onboarding (Phase 2.5): Mounted to existing index + moved to Index directory
# Extraction Loop (Phase 3)
- [ ] [[NoteA (Concept)]] + Luhmann Scan
- [ ] [[NoteB (Concept)]] + Luhmann Scan
# Methodology (Phase 4)
- [ ] [[ToolA (Method)]] (SOP/Checklist/MVE)
# Review (Phase 5 & 6)
- [ ] Feynman Check (De-jargon check)
- [ ] Network Check (2+ links per note; index onboarding: Inbox or keyword entry; multi-index mount check)
# Workflow Audit (Phase 6.5, mandatory)
- [ ] Invoke **workflow-audit** for execution completeness check (Deming + Gawande perspective), produce audit report, remediate any โ items until all DoD items pass
Definition of Done