Codebase Stats

Count lines of code with auto-detected project structure. No external tools required.

Tools to use: Bash for file discovery and counting, Glob and Read for config file detection. Do NOT use cloc , tokei , scc , or any external LOC counter.

Activation

Trigger on:

• /codebase-stats or /loc

• "count lines of code", "LOC by module", "code stats", "project size"

Execution Flow

Step 1: Detect Project Structure

Use Glob and Read to inspect config files at the project root. Follow this decision tree in order (first match wins):

1a. JS/TS Workspace Monorepo

Signal files: pnpm-workspace.yaml , turbo.json , nx.json , lerna.json , or root package.json containing "workspaces"

Module extraction:

• pnpm-workspace.yaml : read packages globs, expand with Glob

• package.json workspaces: read array, expand globs

• Each directory with its own package.json = one module

• Label: workspace

1b. Rust Workspace

Signal: root Cargo.toml containing [workspace]

Module extraction: parse members array, expand globs. Label: workspace

1c. Java / Kotlin Multi-Module

Signal: root pom.xml with <modules> , or settings.gradle / settings.gradle.kts

Module extraction:

• Maven: parse <module> tags

• Gradle: parse include(...) or include '...' , convert :a:b to a/b

• Label: maven-module or gradle-module

1d. Go Multi-Module

Signal: go.mod found in 2+ directories

Module extraction: each directory containing go.mod = one module. Label: go-module

1e. Python Multi-Package

Signal: pyproject.toml or setup.py found in 2+ directories

Module extraction: each directory containing either file = one module. Label: py-package

1f. Full-Stack Monolith

Signal: top-level directory pairs matching known patterns:

• src

• src-tauri (Tauri)

• frontend

• backend

• client

• server

• web

• api

• app

• server

Module extraction: each matched directory = one module. Label: monolith

1g. Simple Single-Module

Fallback: none of the above matched.

Module: project root as the only module. Label: single

Always: add a <root> pseudo-module for files in the project root not belonging to any detected module.

Step 2: Count Lines

Run a single Bash script. The counting logic:

Template - adapt ROOT, MODULES, PRUNE_DIRS as needed

find "$ROOT"
( -name .git -o -name node_modules -o -name target -o -name dist
-o -name build -o -name out -o -name coverage -o -name .next
-o -name .nuxt -o -name .svelte-kit -o -name .turbo -o -name .cache
-o -name pycache -o -name .venv -o -name venv -o -name .mypy_cache
-o -name .pytest_cache -o -name vendor -o -name third_party
-o -name .yarn -o -name .pnpm-store -o -name Pods -o -name DerivedData
-o -name .gradle -o -name .idea -o -name .vscode -o -name .settings
-o -name gen -o -name generated
-o -name bin -o -name obj -o -name tmp -o -name temp
) -prune -o
-type f ! -name '.min.js' ! -name '.min.css' ! -name '.bundle.js'
! -name '.map' ! -name '.snap' ! -name '.lock' ! -name '.d.ts'
! -name 'package-lock.json' ! -name 'pnpm-lock.yaml' ! -name 'go.sum'
! -name '.pb.go' ! -name '_pb2.py'
! -name '.generated.' ! -name '.gen.' ! -name '.g.dart'
! -name '.designer.'
-print0 |
while IFS= read -r -d '' f; do

skip symlinks

[ -L "$f" ] && continue

skip binary

grep -qI . "$f" 2>/dev/null || continue

get extension or basename for extensionless files

base=$(basename "$f") case "$base" in Makefile|Dockerfile|Vagrantfile|Rakefile|Gemfile|Justfile|CMakeLists.txt) ext="$base" ;; .) ext="${base##*.}" ;; *) ext="$base" ;; esac

count

lines=$(wc -l < "$f" 2>/dev/null) || continue loc=$(grep -cve '^[[:space:]]*$' "$f" 2>/dev/null || echo 0)

output: path<TAB>ext<TAB>lines<TAB>loc

printf '%s\t%s\t%d\t%d\n' "$f" "$ext" "$lines" "$loc" done

File type mapping: Apply this extension-to-type table to classify each file:

Type Extensions

ts .ts

tsx .tsx

js .js , .mjs , .cjs

jsx .jsx

py .py

rs .rs

go .go

java .java

kt .kt , .kts

swift .swift

c .c

cpp .cc , .cpp , .cxx

h .h , .hh , .hpp , .hxx

css .css , .scss , .sass , .less

html .html , .htm

sql .sql

sh .sh , .bash , .zsh , .fish

toml .toml

json .json , .jsonc

yaml .yml , .yaml

xml .xml

md .md , .mdx

vue .vue

svelte .svelte

dart .dart

rb .rb

php .php

cs .cs

scala .scala

elixir .ex , .exs

docker Dockerfile

make Makefile , CMakeLists.txt

Unmatched extensions: group into other . Skip known non-source: .png , .jpg , .jpeg , .gif , .svg , .ico , .pdf , .woff , .woff2 , .ttf , .eot , .mp3 , .mp4 , .webm , .webp , .zip , .tar , .gz , .dmg , .exe , .dll , .so , .dylib , .DS_Store .

Module assignment: for each file path, find the longest-matching module root prefix. Files not matching any module go to <root> .

Step 3: Render Output

Language: Match the user's prompt language. If the user writes in Chinese, use Chinese table headers and labels (e.g., "扫描根目录", "文件类型", "模块", "占比"). If in English, use English. Technical terms (module names, file types, paths) stay as-is regardless of language.

Produce four Markdown tables in this exact order:

Table 1: Project Summary

Table 2: By File Type

Sorted by LOC descending.

Share = LOC / total LOC * 100 , 1 decimal place.

Table 3: By Module

Sorted by LOC descending.

Top Types : top 3 file types by LOC within the module, formatted as type(Xk) using abbreviation (e.g., 1.2k for 1,200).

Table 4: Module x Type Detail (Top 20)

Only show if multiple modules detected. Sorted by LOC descending, capped at 20 rows.

If more than 20 combinations exist, append: Showing top 20 of N combinations.

Number Formatting

• Use thousand separators for numbers >= 1,000 (e.g., 59,040 )

• Percentages: 1 decimal place (e.g., 31.2% )

• Abbreviated form in Top Types: 1.2k for 1,200

Edge Cases

Case Behavior

Empty project Show summary with zeros, note "No source files found"

Symlinks Never follow, count in "Skipped"

Binary files with source extension Detect via grep -qI , skip, count in "Skipped"

Very large repo (>10k files) Print warning before counting, continue normally

Permission errors Skip file, increment "Skipped" count

Nested monorepo (workspace inside workspace) Use deepest module root (longest prefix match)

Generated directories (gen/ , generated/ ) Exclude by default

other share > 15% Warn user to review unmapped extensions

Performance Tips

• Always use find -prune to skip heavy directories early

• Process files in a single find | while read loop, not per-file tool calls

• Aggregate with awk or sort | uniq -c , not per-row processing

• Use Read only for config files (manifests), never for source files

• For repos with >5,000 files, run counting in a single Bash invocation