Orbitant Learn

Overview

The core learning experience. Start a new lesson from the backlog, resume the current one, or pick up a postponed lesson. Deliver content adapted to the user's chosen style, track observations silently, and pace to the user's session length.

When to Use

• User wants to start a learning session
• User wants to continue where they left off
• User has free time and wants to learn

Prerequisites

Profile must exist at ~/.claude/learning/profile.md. If not, tell the user: "No learning profile found. Run /init-learning first to set up your profile."

Process

1. Load State

Read these files:

• ~/.claude/learning/profile.md — style preference, session_length
• ~/.claude/learning/lessons/current.md — active lesson (may not exist)
• ~/.claude/learning/backlog.md — available items
• Scan ~/.claude/learning/lessons/archive/ for lessons with status: postponed

2. Route

If current lesson exists (current.md is present):

• Show: "You have an active lesson: [topic] — milestone X/Y. Last session: [date]."
• Options: Continue / Postpone / Abandon
• If Postpone or Abandon: trigger the /end-learn closing flow (ask reason if abandoning), then proceed to "no current lesson" flow below

If no current lesson:

• List postponed lessons (if any): "You have postponed lessons you can resume:"
• List backlog items: "Your backlog:"
• User picks one
• If postponed lesson picked: move it from archive/ back to lessons/current.md, resume from last milestone
• If backlog item picked: expand into a full lesson — generate milestones fitted to session_length, create lessons/current.md

2.5. Build Learner Context

Invoke the learning-insights skill in "build context" mode. It reads archived lessons, session logs, and insights.md to produce a learner context.

If fuzzy_topics found, flag them to the user: "Before we start — these came up as fuzzy in past sessions:

• [topic] (from [lesson], [timeframe ago])
• [topic] (from [lesson], [timeframe ago]) Want to spend a few minutes on any of these, or dive in?"

Wait for the user's choice. If they want to review, spend a few minutes on the topic (a quick question or explanation), then proceed to the lesson. If they want to dive in, proceed.

If cross_topic_patterns found, mention briefly: "Heads up: I've noticed [theme] keeps coming up across different topics. I'll keep an eye on that today."

This is informational only — do not wait for a response, proceed to Session Start.

Difficulty adjustment feeds into milestone generation in step 4:

• If the trend says "advanced" for this gap: skip beginner scaffolding, present harder challenges
• If "beginner": provide more guidance, smaller steps
• If no data yet: use default difficulty (intermediate)

3. Session Start

Ask: "Hands-on or reading today?" (show the default from profile, user can override)

If hands-on, also ask: "Where should we create the project?" (suggest a path, user confirms)

• Record project_path in current.md

4. Deliver Content

Adapt to the chosen style for this session:

Challenge-based:

• Present the milestone as a problem statement with clear success criteria
• Let the user work — do not write the solution
• Give hints only when the user asks or is clearly stuck
• Review the user's code/work, point out issues, suggest improvements
• When the milestone is met, confirm and move to the next

Example challenge prompt:

**Milestone 2: Fetch real API data**

Your task: Replace the mock data with real API calls.

**Success criteria:**
- App fetches data from the /api/metrics endpoint on mount
- Shows a loading spinner while fetching
- Displays an error message if the request fails
- Table renders with real data on success

**Hints available** — just ask if you get stuck. Try it yourself first!

Guided:

• Walk through the milestone step by step
• Explain concepts as you go — WHY, not just how
• Write some code, then ask the user to write the next piece
• More scaffolding than challenge-based, but still ask the user to do the work
• After each step, check understanding: "Makes sense? Any questions before we move on?"

Reading:

• Curate relevant articles, documentation, or book chapters (use web search for up-to-date content)
• Explain key concepts in your own words — concise, not lecture-style
• After each concept block, ask a comprehension question to test understanding
• Propose thought exercises (no coding required)
• This mode works on mobile — keep responses readable, use short paragraphs, avoid code blocks when possible

5. Silent Observation Tracking

Throughout the session, silently track these observations (do NOT show them to the user during the session — they are for /end-learn):

• Hints given: what the user asked for help with, and what hint you provided
• Struggles: where the user got stuck, made repeated mistakes, or needed multiple attempts
• Nailed it: things the user got right on the first try
• Aha moments: when the user demonstrated sudden understanding
• Questions asked: what the user was curious about

Keep these observations in your conversation context as a mental running list. They will be synthesized by /end-learn. Do NOT write them to any file during the session.

Example internal observation (never shown to user):

Observation: Hint given — user confused useState return value, explained it's [value, setter].
Observation: Nailed it — destructuring props on first try without help.
Observation: Struggle — tried to mutate state directly 3 times before understanding immutability.

6. Time Awareness

• Track approximate elapsed time based on conversation flow
• When approaching the user's session_length: "You're at about ~Xm — good stopping point after this exercise, or keep going?"
• Never cut the user off — always let them choose to continue
• If a milestone completes near the time limit, suggest ending: "Milestone done! Great stopping point."

7. Session End

When the user wants to stop (or you suggest stopping and they agree):

• Tell the user: "Run /end-learn to save your progress and capture what you learned."
• Do NOT write files yourself — /end-learn handles all persistence

Milestone Design (when expanding from backlog)

When creating a new lesson from a backlog item:

• Each milestone should be completable in one session (user's session_length)
• Milestones build on each other — each one produces a visible, testable result
• First milestone is always the simplest — get something working immediately
• Last milestone adds polish or advanced features
• For reading lessons: milestones are topic chunks, not code deliverables

Important

• ONE lesson at a time — never create a second current.md
• Observation tracking is SILENT — do not show the user your notes mid-session
• Respect the user's style choice — do not switch styles mid-session unless asked
• If the backlog is empty and no lessons exist, suggest running /new-learning
• Preserve current.md session log between sessions — append, don't overwrite