repo-understand

# Purpose Create a reliable curriculum map before deep teaching. Prevents shallow file retrieval by forcing the agent to infer learning structure, identify prerequisite dependencies, and recommend coherent starting paths. # Activation - User asks to learn a repository, course repo, or technical curriculum. User asks "where should I start?" in a multi-module repo. Agent hasn't established module sequence and prerequisites. - **Skip if**: user wants a narrow operational task unrelated to learning flow, or repo context is absent and user wants generic theory. - **Routing**: run before `teach-concept`, `socratic-mode`, `check-understanding`, and `build-with-me` for repo-based learning. # Inputs - Repository root and visible structure, high-signal docs (README, syllabus, indexes), learner goal and level signals. # Workflow 1. **Classify** — Determine repo type: curriculum-first, project-first, reference-first, or hybrid. Identify learning artifacts (lessons, labs, projects, checkpoints). 2. **Map Hierarchy** — Derive module/phase order from folder structure and docs. Produce a phase map with outcomes per phase. 3. **Infer Prerequisites** — For each phase, list conceptual and tooling prerequisites. Mark dependency edges between phases. 4. **Estimate Progression** — Foundational → applied → advanced. Identify transition points where learners often stall. 5. **Synthesize Paths** — Propose 2–3 path variants based on learner goals and level, with tradeoffs and estimated effort. 6. **Checkpoint** — Ask one decision-forcing question to select a path. Do not start deep instruction before the learner selects or confirms. # Rules - DO: require a phase map before any deep explanation. - DO: explain why sequence matters, not just what modules exist. - DO: provide goal-conditioned paths with tradeoffs — never a single path for all learners. - DO: surface uncertainty explicitly when repo signals are weak. - DON'T: quote random files without structural context. - DON'T: recommend advanced topics before prerequisite chain is visible. - DON'T: dump long directory trees — cap orientation to phase outcomes and decisions. - DON'T: invent lessons/modules without evidence — label inferred items as tentative. # Output Responses should contain: repo type classification, learner goal/level, curriculum map (phases + outcomes), prerequisite chain, 2–3 recommended paths with tradeoffs, checkpoint question, and next step. Format naturally. # Checklist - [ ] Curriculum map includes ordered phases and outcomes. - [ ] Prerequisite chain is explicit. - [ ] At least two goal-conditioned paths provided. - [ ] Checkpoint question asked before deep teaching begins.