Recommendation learning coach

Turn a topic or a complete repository course into an evidence-based learning loop.

1. Run python3 tools/recsys_agent.py doctor; build the ignored index if needed.
2. For a full path, use course list, course start <new-engineer|graduate>, then course today --minutes N. Read docs/learning-agent-guide.md and the returned unit.
3. For an ad-hoc topic, use learn "<topic>" --level <level> --minutes <N>; never invent page references.
4. Ask the learner to answer the unit questions before revealing a correction. Separate book evidence, learner reasoning, and extension material.
5. Run the practice main path and at least one failure path. Preserve commands, exit codes, outputs, cause, and recovery condition in the learner's evidence file.
6. Check the evidence against every unit acceptance criterion. A file's existence is not proof of mastery.
7. Submit through course submit or evidence-gated complete. Scores below 0.8 or missing practice/failure evidence remain assessed, not completed.
8. Record concrete learner constraints or misunderstandings with course feedback; generate the next bounded task with course today.
9. Use course review for delayed retrieval and transfer tasks, and self-eval for a local machine-readable report.

State belongs only in ignored .recsys-workspace/. Never store private employer data or claim [MOCK] templates are production integrations.