feat: mentor skill

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

mentor/README.md

@@ -0,0 +1,24 @@
+# Mentor path
+
+Take the course with an AI agent as a Socratic mentor. The agent asks, you build: it never writes your engine or content, releases hints one tier at a time, and holds you to the Zero-Coupling Law with the course's tools.
+
+`SKILL.md` is an [Agent Skill](https://docs.claude.com/en/docs/claude-code/skills) (Claude Code compatible). It contains only teaching rules; the curriculum is read from the course repository.
+
+## Install
+
+```sh
+mkdir -p ~/.claude/skills
+cp -r mentor ~/.claude/skills/systemic-design-mentor
+```
+
+## Use
+
+1. Clone this repository. Copy `workspace-template/<rust|ts>` to a directory of your own and put it under git (see Module 0).
+2. Open your agent in **your workspace**, not the course repository, and say: "start the systemic design course".
+3. The mentor asks once for the course repository path and stores it in the workspace file `.course-progress.json`:
+   `{ "courseRoot": "...", "module": "04", "beat": "Build", "hintsUsed": 1, "expansionBase": "<git ref>" }`.
+   Add it to your `.gitignore`.
+
+The mentor runs `tools/start-module.mjs`, `tools/coupling-audit.mjs` and `tools/expansion-check.mjs` from the course repository against your workspace. It may read and test your code but will not edit `engineDirs`, `contentDirs` or the harness. Ask it "does the real game do X" and it answers only from the lesson's cited sources, or says the course does not cover it.
+
+Module 0 uses `naive-engine/` rather than the workspace, so no tests are copied for it.