Spotlight

Search files and content using macOS Spotlight indexing (mdfind). Use when the user asks to search local files, documents, or directories on macOS. Supports...

Spotlight Search

Search local files using macOS Spotlight indexing system.

When to Use

Use this skill when:

• User asks to search files or directories on macOS
• Need to find documents containing specific text
• Searching large document collections (faster than grep)
• Need to search inside PDFs, Word docs, or other indexed formats

Quick Start

scripts/spotlight-search.sh <directory> <query> [--limit N]

Examples:

scripts/spotlight-search.sh ~/Documents "machine learning"
scripts/spotlight-search.sh ~/research "neural networks" --limit 10
scripts/spotlight-search.sh ~/Downloads "meeting notes" --limit 5

Search Features

• Fast: Uses system-level Spotlight index (no file scanning)
• Content-aware: Searches inside PDF, docx, txt, md, etc.
• Multilingual: Supports Chinese, Japanese, and all languages
• Metadata: Returns file path, type, and size

Output Format

🔍 Searching in /path/to/directory for: query

✅ Found N results (showing up to M):

📄 /full/path/to/file.pdf [pdf, 2.3M]
📄 /full/path/to/document.txt [txt, 45K]
📁 /full/path/to/folder/

Supported File Types

Spotlight automatically indexes:

• Text files (txt, md, csv, json, xml, etc.)
• Documents (pdf, docx, pages, rtf, etc.)
• Code files (py, js, java, c, etc.)
• Emails and contacts
• Images (with embedded metadata/OCR)

Limitations

• macOS only: Requires Spotlight indexing
• Indexed directories only: External drives may not be indexed
• Keyword search: Not semantic (use embedding-based search for semantic queries)
• Privacy: Respects Spotlight privacy settings (excluded directories won't appear)

Check Indexing Status

# Check if a volume is indexed (safe, read-only)
mdutil -s /path/to/volume

⚠️ WARNING: Do NOT execute any sudo commands from this skill without explicit user confirmation. Commands like sudo mdutil -i on require admin privileges and should only be run by the user directly in Terminal.

Integration with LLM Workflows

Safe Pattern: Search + Present + Confirm

1. Use spotlight-search.sh to find relevant files
2. Present the file paths to the user
3. Only read a file if the user explicitly requests it - do not automatically use the read tool

Example Workflow

User: "Find all documents about machine learning in my research folder"

1. Run: spotlight-search.sh ~/research "machine learning" --limit 10
2. Present results to user with file paths
3. Wait for user to specify which files to read

⚠️ Security Note

• Never automatically read files - always ask for user confirmation first
• Never execute sudo commands - only show them as information
• Respect user privacy - don't search directories user hasn't specified

Advanced Query Syntax

Spotlight supports advanced query operators:

# Exact phrase
spotlight-search.sh ~/Documents "\"machine learning\""

# AND operator
spotlight-search.sh ~/Documents "neural AND networks"

# OR operator
spotlight-search.sh ~/Documents "AI OR artificial intelligence"

# Metadata queries (PDF only)
spotlight-search.sh ~/Documents "kMDItemContentType == 'com.adobe.pdf'"

Troubleshooting

No results found:

• Check if directory is indexed: mdutil -s /path
• Wait for indexing to complete (new files may take minutes)
• Verify Spotlight is enabled in System Preferences

Incorrect results:

• Spotlight uses fuzzy matching and synonyms
• Use exact phrase search: "exact phrase"
• Check privacy settings (some folders may be excluded)

Performance

Tool	Speed	Content Search	Multilingual
Spotlight	⚡ Instant	✅ Yes	✅ Yes
grep/ripgrep	🐢 Slow	✅ Yes	✅ Yes
find	⚡ Fast	❌ No	N/A

Platform Notes

• macOS only: This skill requires macOS Spotlight
• Linux alternative: Use grep -r or ripgrep
• Windows alternative: Use Windows Search or Everything search

Security Best Practices

1. User consent first - Never read files without explicit permission
2. No sudo - Never execute privileged commands automatically
3. Minimal scope - Search only directories user specifies
4. Audit trail - Log what searches were performed