cc-self-train adaptive learning system: a three-layer hook-based engagement observation and scaffolding adjustment mechanism¶
Design · All designs
Evidence · no claims cited
Description¶
The adaptive learning system observes learner engagement quality through keyword-heuristic classification of each learner message into six engagement categories with quality scores from 1 to 5, accumulated in a local learner profile. Layer 1 is a silent Stop hook that fires after every Claude response and computes streak booleans; Layer 2 injects a teaching note into Claude's context at session start based on the productive/unproductive ratio; Layer 3 adjusts the learner's Effective Level at module boundaries. The article describes "a two-timescale system" in which a slow clock adjusts the persona schedule at module boundaries using aggregate statistics while a fast clock triggers scaffolding changes mid-module using sequential patterns, governed by an asymmetric response principle that is quicker to increase scaffolding than to withdraw it.
Design Implications¶
Context¶
Requirements¶
- Operates entirely through Claude Code's hook system with no external dependencies, no API calls for observation, and no web backend
Constraints¶
- Observation uses keyword heuristics rather than LLM-based classification, so individual classifications are noisy; Effective Level changes occur only at module boundaries; level changes are silent to the student
Target Learners¶
- developers learning an agentic AI coding tool through the cc-self-train curriculum
Learning Goals¶
- sustained productive engagement and appropriately paced scaffolding during tool-mastery learning
Claims¶
Related Designs¶
- Closed-loop three-layer interpretable ITS architecture linking multimodal scoring, expressive diagnostics, and conversational coaching
- HeuristicEdu: an SFT-then-GRPO alignment pipeline with a three-component heuristic reward for Socratic tutoring
- TANDEM radar-tracking simulation training task with 2x2x2 fully crossed experimental design (N = 350)
- cc-self-train auto-updating sync pipeline — how the course's modules are kept in step with the tool they teach
Examples¶
-
Key Sources¶
- Zain Naboulsi. (2026). Agentic Education: Using Claude Code to Teach Claude Code. arXiv. https://arxiv.org/abs/2604.17460
Page metadata
| Field | Value |
|---|---|
type |
design |
id |
cc-self-train-adaptive-learning-system |
title |
cc-self-train adaptive learning system: a three-layer hook-based engagement observation and scaffolding adjustment mechanism |
description |
The adaptive learning system observes learner engagement quality through keyword-heuristic classification of each learner message into six engagement categories with quality scores from 1 to 5, accumulated in a local... |
status |
review |
generated |
by process:settle-candidates on 2026-10-10 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — zain-naboulsi-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.