Repository Index Creator

Create or update PROJECT_INDEX.md — a compact markdown index that gives the LLM full codebase navigation without reading every file.

Why

• Reading all files costs ~58,000 tokens per session
• Reading the index costs ~1,000 tokens (98% reduction)
• Exports in the index save additional Read tool calls during discovery

How to Run

Create index: /index-repo Update index: /index-repo update

Process

1. Glob all source and test files in parallel:

   • src/**/*.py, tests/**/*.py
   • *.toml, Makefile

2. Read each file, extract:

   • Purpose (1-line description)
   • Exports (public classes, functions, constants)
   • Status (only mark stubs — implemented is the default)
   • Endpoints (for routers)
   • Fixtures (for test conftest files)

3. Write PROJECT_INDEX.md using the compact format below.

Output Format

Single file: PROJECT_INDEX.md (target: <3KB)

# {Project Name} | {Framework} | {Python version} | {Package manager}

## Commands
run: {command}
test: {command}
lint: {command}
fix: {command}

## Architecture
- key: value (one line per architectural decision)

## Modules

path/to/file.py — one-line purpose
  exports: Class1, Class2, function1

### path/to/package/ [STUB if not implemented]

file.py — purpose
  exports: ExportedName

## Tests ({count} total, {status})

conftest.py — fixtures: fixture1, fixture2
test_file.py ({count}), test_other.py ({count})

Rules

• One line per module: path — purpose
• Exports on indented line below: exports: Name1, Name2
• Only mark exceptions: stubs get [STUB], implemented is assumed
• Group by directory with ### headers to avoid repeating paths
• Tests are compact: test_file.py (count) on shared lines
• No duplicated data: skip deps (in pyproject.toml), CI config, test class names
• Keep under 3KB