# Reference run

The authoring assistant executed the starter and repair locally with Python 3.14 on Linux. This records a command-line check of the supplied files, not a recorded Claude Code session or a VS Code screen demonstration.

## Before repair

From `starter/`:

```bash
python3 build_index.py
python3 -m unittest -v
```

The builder reported `Wrote INDEX.md`; the index contained only Welcome.
The test command exited 1: five tests ran, with `test_nested_note_is_linked` failing and the others passing.

The failing assertion was:

```text
AssertionError: '[Agent loops](notes/research/loops.md)' not found in '# Notes index\n\n'
```

## Repair and result

The reference implementation changes `notes.glob("*.md")` to `notes.rglob("*.md")` in `build_index.py`.

From `solution/`:

```bash
python3 -m unittest -v
python3 build_index.py
```

The test command exited 0 with five passing tests. The generated index linked Agent loops, Persistent memory and Welcome. A separate filesystem check resolved every generated link to its sample file. The test source and sample note bytes matched the starter after the repair.

## What still needs a learner run

VS Code interface instructions were checked against its official documentation; the GUI was not exercised here. Windows and macOS commands were not executed here. The fresh-session handover prompts have not been tested with a separate agent. The learner run should verify that the agent reads the handover, reruns the checks and accurately describes the existing project without edits.

This is a basic Markdown-note fixture. It does not claim to index every Markdown dialect, filesystem layout or Obsidian feature.