Identify layout smells

Identify Layout Smells

Analyze file and directory organization to find structural problems that make a codebase hard to navigate for humans and AI. These smells arise when code is split to meet mechanical constraints (line limits, linter rules) without restructuring the module boundaries.

The core principle: file system layout should communicate domain structure at a glance, without reading the code.

Smell Catalog (quick reference)

#	Smell	One-line test
1	Prefix Cluster	3+ files share a name prefix in a flat directory
2	Shim Facade	File primarily re-exports symbols from siblings
3	Junk Drawer	File named core/utils/helpers/common/misc with unrelated contents
4	Import Magnet	File imported by >50% of siblings AND mixes responsibilities
5	Shadow Name	File named by relationship (_helpers, _utils) not by content
6	Orphan Module	File imported by nothing (excluding entry points, tests)
7	Layering Violation	Flat directory mixes high-level orchestration with low-level infra
8	Infrastructure Sprawl	5+ infra files (transport, config, protocol, caching) ungrouped in a flat directory
9	Misplaced Module	>70% of a file's importers are in one package, but the file lives outside it
10	Coupled Pair	Two files with a clear parent/satellite relationship not grouped into a package
11	Import-Tree Cluster	A root file + 5 or more descendants all flat in one directory (hidden feature tree)
12	Premature Package	A directory containing only 1 source file

For detailed detection criteria, examples, and remediation per smell, consult references/smell-catalog.md.

Analysis Workflow

Step 1: Map the directory

List all source files in the target directory (non-test, non-config). Record filenames and line counts. This is the raw material.

# Python
find <dir> -maxdepth 1 -name "*.py" ! -name "__init__.py" ! -name "conftest.py" | sort

# TypeScript
find <dir> -maxdepth 1 \( -name "*.ts" -o -name "*.tsx" \) ! -name "index.ts" | sort

Step 2: Detect prefix clusters and import-tree clusters

Name-based: Group filenames by shared prefix (4+ chars, 3+ files). Exclude test_/.test./.spec.. Report each cluster with member files and total line count. A cluster with high total lines (>500) is a strong signal for package promotion.

Import-tree-based: For component directories (React, Vue, etc.), trace import trees from root-level files. If a root file imports 3+ siblings, and those siblings import further siblings, count the entire tree. If the tree totals 5+ files all flat in one directory, that's an import-tree cluster — a hidden feature that should be a subdirectory. This catches functional groupings that name-based detection misses (e.g., ChatApp + ChatInput

• ChatTimeline + ContextBar + ContextViewer + ReportDialog form a chat feature tree, but only 3 share the Chat prefix).

Step 3: Scan for junk drawers and shadow names

Check filenames against known patterns:

• Junk drawer names: core, utils, helpers, common, shared, misc, lib, base
• Shadow suffixes: _helpers, _utils, _data, _support, _extra, _impl
• Exclude _types/_schemas in TypeScript (convention), _test/_spec (tests)

For each match, read the file and list its top-level definitions to assess whether the name hides what's actually inside.

Step 4: Build the import graph

For each source file, extract its imports from sibling files. Produce an adjacency list: file -> [siblings it imports from].

# Python: use ast.parse to extract ImportFrom nodes
# TypeScript: grep for import statements with relative paths

Use this graph to detect:

• Import magnets — files imported by >50% of siblings (cross-check: do they mix responsibilities?)
• Orphan modules — files with zero incoming edges (excluding entry points)
• Shim facades — files where >50% of top-level symbols are re-exports
• Layering violations — compute depth (leaf = imported by many, imports few; root = imports many, imported by few). If 3+ layers coexist in one flat directory, flag it.
• Infrastructure sprawl — classify by content, not just degree. If a file contains no domain logic (only transport, config, protocol, caching, tracing), it's infra regardless of in/out-degree. If 5+ such files sit ungrouped in the same flat directory, they should be a package. This smell survives prefix-cluster fixes because each file has a unique name.
• Misplaced modules — for each file outside a package, count what percentage of its importers live in a single package. If >70% come from one package, the file probably belongs in that package.
• Coupled pairs — two files with a clear parent/satellite relationship (e.g., search.py + search_queries.py, entities.py + entity_schemas.py) that aren't grouped into a subdirectory. Report as LOW severity — not actionable alone, but worth noting when multiple pairs coexist in the same directory.
• Premature packages — directories containing only 1 source file. The directory adds navigation overhead without providing grouping value. Flag as LOW; the fix is either to add more files or inline the file into the parent directory.
• Dev/production layering — when checking for layering violations, also look for dev-only files (log viewers, test harnesses, debug panels, storybook files) mixed into production component directories. Dev tools scattered across directories are a special case of misplaced module — they should be grouped together (e.g., dev/ or components/dev/).

Step 5: Produce the report

Output a structured findings list. For each smell found:

1. Smell name and severity (high/medium/low)
2. Files involved with line counts
3. Evidence — the specific pattern that triggered detection
4. Suggested remediation — concrete rename, move, or package promotion

Group findings by smell type. End with a summary count table.

Severity guidelines

• High: Prefix clusters of 4+ files, import-tree clusters of 7+ files, junk drawers over 200 lines, import magnets with 3+ mixed responsibilities, infrastructure sprawl of 7+ files, misplaced modules with >90% single-package importers
• Medium: Prefix clusters of 3 files, import-tree clusters of 5-6 files, shadow names, shim facades, layering violations, infrastructure sprawl of 5-6 files, misplaced modules with 70-90% single-package importers
• Low: Orphan modules, small junk drawers (<100 lines), single shadow-name files, coupled pairs, premature packages

What NOT to flag

• A settings.py/config.ts imported everywhere is fine (single responsibility, just widely needed)
• A types.ts file co-located with its feature module is convention, not a smell
• Test file prefixes (test_, .test., .spec.) are naming conventions
• __init__.py/index.ts barrel files in a package that already IS a directory
• Small projects (<10 source files) rarely benefit from subdirectories

Additional Resources

Reference Files

• references/smell-catalog.md — Full catalog with detection criteria, language-specific examples (Python + TypeScript), and remediation patterns for all 7 smells