Learn Domain — Feynman-Spine
The Learn domain supplies a Feynman-Spine learning procedure with goal, corpus, explanation, practice and retention artifacts. Skill requests require the selected harness and their actual collaborators; source instructions alone do not certify learning outcomes or credential issuance.
The learning arc
/learn-goal → set what you want to learn
/learn-survey → diagnostic placement + recursion floor
/learn-plan → concept DAG + curriculum builder
/feynman-loop → core Feynman PMPO loop (explain, gap-find, re-study)
/learn-grade → corpus-grounded grading with disclosed screening
/learn-retain → FSRS-6 spaced retrieval
/learn-practice → deliberate practice track
/learn-certify → local self-issued credential procedure
Support skills
| Skill | Purpose |
|---|---|
/learn-kb | KB registry + adapter management |
/learn-about-system | Prometheus stack meta-learning |
/learn-harness | Harness detection + capability map |
/ui-surface | Cross-harness UI rendering primitive |
The optional /sync-status, /sync-peers, and /sync-push skills are
distributed by prometheus-companion, which owns cross-machine replication.
Core invariants
- Self-reported fluency NEVER closes a Feynman loop — all 3 mastery conditions must hold
- Grading needs evidence — sycophancy screening is a bounded diagnostic and does not guarantee correctness or an independent model
- KB destinations are explicit — local reads stay local, while configured Dify/Palace and downstream inference destinations require separate ownership and authorization