index-init

/index-init

Generate a complete project index with per-module README files.

Step 1: Announce

Say: "I'm using the index-init skill to generate a project index."

Step 2: Detect project type

Glob the root directory for these signals. Use first match wins priority:

Priority	Signal	Type	Module detection
1	CMakeLists.txt containing add_subdirectory()	cpp-cmake	Each add_subdirectory() argument = module
2	Cargo.toml containing [workspace]	rust-cargo	Each members entry = crate
3	go.work or multiple go.mod	go	Each use directive or go.mod directory = module
4	package.json containing "workspaces"	node-workspaces	Each workspace pattern match = package
5	*.sln file	dotnet	Each .csproj referenced = project
6	pom.xml containing <modules>	java-maven	Each <module> = subproject
7	build.gradle or settings.gradle containing include	java-gradle	Each include(':...') = subproject
8	pyproject.toml or setup.py	python	Packages or top-level dirs with __init__.py
9	pubspec.yaml	dart-flutter	Top-level or packages/ dirs
10	Fallback	generic	Top-level dirs containing source files (.cpp, .py, .js, .ts, .rs, .go, .java, .cs, .dart)

For each detected type, extract:

• Modules: from build/manifest files as described above
• Purpose: from comments, description fields, docstrings, existing README content
• Dependencies: from link directives (target_link_libraries), imports, manifest deps
• Public API: from interface files (I*.h), exports, pub functions — names + method signatures only
• Files: use git ls-files (respects .gitignore) to list source + config files per module. Fallback to Glob if not a git repo.

Never index: .env, .env.*, .git/, *.key, *.pem, .npmrc, .pypirc, .aws/, .ssh/, *credentials*, *secret*, *token*, binary blobs (images, compiled objects, archives).

Step 3: Generate <Module>/README.md for each module

For each module, create or update <Module>/README.md using this template:

# <Module Name>
<!-- AUTO-GENERATED by /index-init — do not edit above CUSTOM section -->

## Purpose
<1-2 sentences describing what this module does>

## Architecture
<ASCII diagram if inferrable from code structure, otherwise omit this section>

## Public API
- **IClassName** — `method1()`, `method2()`, `method3()`
- **ClassName** — `publicMethod1()`, `publicMethod2()`

## Files
| File | Tags | Description |
|------|------|-------------|
| `relative/path/to/file.ext` | [tag1, tag2] | Short description (<80 chars) |

## Dependencies
- **Internal**: Module1, Module2
- **External**: Library1 v1.2, Library2

## Configs & Scripts
| File | Description |
|------|-------------|
| `CMakeLists.txt` | Build configuration |
| `config.json` | Runtime configuration |

<!-- AUTO-GENERATED END -->

<!-- CUSTOM — your notes below survive re-indexing -->

Rules:

• If README.md already exists and contains a <!-- CUSTOM section, PRESERVE its content entirely
• Generate tags from: file name patterns, class names found in file, function names, import/include keywords
• Generate descriptions from: first comment block, docstring, class/function declarations, or inferred from naming
• Include non-code files: .json, .yaml, .toml, Dockerfile, *.sh, *.ps1, CI config files
• Keep file descriptions under 80 characters

Step 4a: Generate .claude/index.summary.md

This lightweight file is injected at session start. Hard cap: 4KB.

# Project Index (Summary)
<!-- AUTO-GENERATED by /index-init -->
Generated: <YYYY-MM-DD HH:MM> | Type: <project-type> | Modules: <count> | Files: <total-count>

Read `.claude/index.md` for full index with dependency graph and tag index.
Read `<Module>/README.md` for detailed module documentation.

## Modules
| Module | Purpose | Key Tags |
|--------|---------|----------|
| ModuleName | Short purpose (<60 chars) | tag1, tag2, tag3 |
<!-- AUTO-GENERATED END -->

Rules:

• No CUSTOM section (auto-regenerated, never hand-edited)
• Keep module purposes under 60 chars
• If the file would exceed 4KB, truncate the module table with: | ... | N more modules. See .claude/index.md | |

Step 4b: Generate .claude/index.md

Full index file. Claude reads this on demand for details.

# Project Index
<!-- AUTO-GENERATED by /index-init -->
Generated: <YYYY-MM-DD HH:MM> | Type: <project-type> | Modules: <count> | Files: <total-count>

## How to Use
The summary is injected at session start. Read module READMEs for file-level detail.
See CLAUDE.md "Code Navigation" section for the search strategy.

## Modules
| Module | Purpose | Tags | README |
|--------|---------|------|--------|
| ModuleName | Short purpose | [tag1, tag2] | [README](../ModuleName/README.md) |

## Dependency Graph
ModuleA → (foundation)
├── ModuleB → ModuleA
├── ModuleC → ModuleA, ModuleB
└── MainApp → all modules

## Tag Index
tag-name: Module1, Module2
another-tag: Module3, Module4
<!-- AUTO-GENERATED END -->

<!-- CUSTOM -->

Step 5: Update CLAUDE.md

Read the existing CLAUDE.md. If it already has a "Code Index" or "Code Navigation" section, update it. Otherwise, append this section:

## Code Navigation

When you need to search the codebase and don't already know the exact file:

1. Read `.claude/index.md` → Tag Index to find which module(s) own the topic
2. Read `<Module>/README.md` → exact file paths, classes, responsibilities
3. Grep/glob within that module — not the whole repo

Skip this if you already know the file path. This is for discovery, not mandatory ritual.

Index files: `.claude/index.md` (full), `.claude/index.summary.md` (session start), `<Module>/README.md` (per-module). Re-index: `/index-init`

Step 6: Handle --gitignore flag

If $ARGUMENTS contains --gitignore:

• Read .gitignore (create if not exists)
• Add these lines if not already present:

  # MacroManAtlas index files
  .claude/index.md
  .claude/index.summary.md
  

• Do NOT gitignore <Module>/README.md (useful for humans)

Step 7: Completion

Summarize what was generated:

• Project type detected
• Number of modules indexed
• Number of files catalogued
• Remind user that hooks are provided by the plugin (auto-registered via hooks.json)
• Suggest running /index-status to verify