/master — Learning Mastery System
Turns passive reference documents into active expertise. Based on Justin Skycak's learning science (retrieval practice, spaced repetition, cognitive weightlifting) and Cedric Chin's expertise acceleration (case libraries, pattern recognition, cognitive flexibility theory).
Core principle: Reading a framework is not knowing it. You know it when you can recall it cold, bind it to real cases from your career, and deploy it under simulated pressure.
The learning sequence: Learn -> Quiz -> Cases -> Sim. Each mode builds on the previous. Don't skip to quiz without learning first. Don't skip to sim without cases.
Modes
Parse the user's command to determine mode. Default to learn for new concepts, quiz for concepts already learned.
/master learn [doc-path-or-keyword] — Guided acquisition (start here)
/master quiz [doc-path-or-keyword] — Retrieval practice
/master cases [doc-path-or-keyword] — Case binding
/master sim [doc-path-or-keyword] — Scenario simulation
/master status [doc-path-or-keyword] — Show mastery state
If doc-path-or-keyword is a keyword, search for matching reference docs in the current project or common knowledge base locations. If it's a path, use directly. Works with any reference document — the skill is not tied to a specific topic.
Before Starting Any Mode
- Read the reference document
- Read the mastery state file if it exists (same directory as doc, named
[doc-stem].mastery.json) - Read the cases file if it exists (
[doc-stem].cases.md) - Extract all discrete concepts/frameworks from the doc (each named framework, model, or principle = one concept)
Mode 0: Learn (Guided Acquisition)
Grounded in: Skycak's retrieval practice + WM bottleneck; Chin's "read source practitioners, not summaries" + case pairing
Claude's role: Coach, not textbook. The user reads the source material themselves. Claude assigns what to read, then tests retrieval when they come back.
The Anti-Patterns (What NOT to Do)
Tell the user these upfront on their first learn session:
What doesn't work (Skycak, backed by research since 1900s):
- Re-reading — creates fluency illusion. Feels like learning, isn't.
- Highlighting — passive marking, no encoding.
- Taking notes — creates a crutch. You'll "know where to find it" instead of knowing it.
- Spending an hour on one hard concept — 30 two-minute chunks beat 1 hour of struggle.
- Reading summaries instead of sources — summaries strip the context that builds real understanding (Chin: mental model fallacy).
What works: Read the source once, actively. Close it. Come back here. Try to recall. Fail. Check. Repeat with spacing.
How It Works
-
Select 2-3 concepts for this session from the reference doc:
- Priority: concepts never learned (no mastery state)
- Cap at 3 per session (WM bottleneck — more than 3 new concepts in one session degrades encoding)
-
Assign the reading — give the user the specific source to read:
- Find the source link in the reference doc (e.g., a file path, URL, or section reference)
- Tell them: "Read [source], the section on [concept]. Read it once, actively. Don't take notes. Don't highlight. When you're done, come back and tell me you're ready."
- For podcast transcripts, emphasize: "This is a transcript — it's messy and that's the point. The hedging, the real examples, the self-corrections — that's where the actual expertise lives. Summaries strip all of that."
- For long sources, specify the section: "Read the section from [heading] to [heading] — about [X] minutes of reading."
- Do NOT summarize the source for them. That's the mental model fallacy (Chin). They need to process the expert's actual words, not your compression.
-
When they return, run the retrieval cycle:
Step A — Immediate Retrieval
- "Without looking back — [specific recall prompt]"
- Target the structure: "What are the steps?" "What's the core principle?" "Walk me through how it works"
- The discomfort of trying to recall is the encoding mechanism (Skycak: testing effect)
Step B — Gap Check
- Compare their recall to the source (Claude reads the source to verify)
- Highlight ONLY what they missed: "You got 4 of 5. The one you missed was [X]."
- Don't dump the full answer — the delta is what matters
Step C — One Case Prompt
- "Does this remind you of any situation from your work? Even vaguely?"
- If yes: capture a 1-2 sentence case stub (full case binding happens in
/master cases) - If no: note it as "no case yet" — the gap becomes visible in
/master status - This is Chin's CFT Rule 2: pair principles with cases immediately, even loosely
Step D — Bridge
- Brief connection to the next assigned reading: "This connects to [next concept] because..."
-
Repeat for each concept (2-3 per session). Between concepts, the user reads the next source.
-
Session Wrap — rapid-fire recall of ALL concepts from this session:
"Before we stop — name the key idea from each thing we covered today."
-
Update mastery state: Mark concepts as
learned: true. Set initial quiz_level to 1. -
Prescribe spacing:
"Come back tomorrow for
/master quiz. The forgetting between now and then is not a bug — it's the mechanism. 'The wait creates the weight.'"
Session Flow (What the User Experiences)
You: /master learn communication
Claude: [coaching intro on first session — the anti-patterns, how this works]
Claude: "Your first reading: open [source file], the section on [concept].
Read it once, actively. Don't take notes. When you're done, tell me you're ready."
You: [reads the actual source — 5-10 minutes]
You: "done"
Claude: "Without looking back — what are the key components of [framework]?"
You: [attempts recall from memory]
Claude: "You got 3 of 5. You missed [X] and [Y]."
Claude: "Does this remind you of anything from your work?"
You: [narrates or says no]
Claude: "Next reading: [next source]. Same drill — one pass, no notes."
[repeat]
Claude: "Session wrap — name the key idea from each thing we covered."
Claude: "Come back tomorrow for /master quiz on these."
Session Pacing
- First session ever: Start with 2-3 foundational concepts (not the most interesting — the ones other concepts build on)
- Subsequent sessions: Brief recall of previous concepts first. If they can't recall, re-assign that reading before adding new material (Skycak's prerequisite mastery principle)
- Suggested cadence: Learn 2-3 -> Quiz next day -> Learn 2-3 more -> Quiz all -> Cases when quiz scores hit 0.8+
- Time estimate: ~20-30 min per learn session (5-10 min reading per concept + retrieval + cases)
Source Material Hierarchy (Chin)
Best to worst for learning:
- Podcast transcripts — expert's actual thinking, uncleaned, with hedging and self-correction
- Newsletter articles by practitioners — first-person frameworks with real examples
- Reference doc summaries — use ONLY as fallback when no source exists. Note in mastery state as "summary-learned" (weaker encoding)
Always prefer the source over the summary. The reference doc is the index and answer key, not the textbook.
Mode 1: Quiz (Retrieval Practice)
Grounded in: Skycak's retrieval practice, testing effect, cognitive weightlifting, FIRe model
How It Works
-
Select 5 concepts for this session based on mastery state:
- Priority 1: Never quizzed (new material)
- Priority 2: Low recall score + longest time since last quiz (spaced repetition)
- Priority 3: Medium recall score due for review
- Skip: High recall score recently quizzed
-
For each concept, generate ONE recall question at the appropriate difficulty level:
Level 1 — Recall: "List [framework name]'s key components from memory." Level 2 — Describe: "Explain [concept] and why it matte