luum-home/luum-cognitive-os/detect-patterns

Purpose

Systematically finds code-level rot that accumulates over time. Detects four categories of systemic problems:

1. Dead Metadata — frontmatter fields in SKILL.md files that no code ever reads
2. Broken Chains — imports or references to files that don't exist (symlink-aware)
3. Phantom Entries — catalog/config entries pointing to non-existent components
4. Structural Tests — tests that only check file existence, never exercise behavior

Invocation

/detect-patterns [--type dead-metadata|broken-chains|phantoms|structural-tests|all]

Default: --type all

What to Do

Step 1: Run the pattern detector

from cos_lib.pattern_detector import PatternDetector
detector = PatternDetector()
# Run all detectors or a specific one
if type_arg == "all":
    results = detector.run_all(project_dir)
else:
    results = detector.run_type(project_dir, type_arg)

Step 2: Format and present results

report = detector.format_report(results)
print(report)

Step 3: Summarize actionable items

Group findings by severity and present:

• CRITICAL — broken chains that will cause runtime errors. Fix immediately.
• WARNING — dead metadata and phantom entries. Fix during next cleanup.
• INFO — structural tests. Improve when touching those test files.

Step 4: Suggest next steps

• For broken chains: identify the source and target, suggest creating or fixing the reference
• For dead metadata: suggest removing unused frontmatter keys
• For phantoms: suggest creating the missing component or removing the catalog entry
• For structural tests: suggest adding behavioral assertions

Rules

• ALWAYS resolve symlinks with os.path.realpath() before checking existence
• ALWAYS handle try/except ImportError gracefully when checking Python imports
• When checking config flags, grep for the YAML key in .py and .sh files
• Report evidence: show what was checked and what was found/not found
• Do NOT auto-fix — report only, let the human decide