| name | Large File Handling |
| description | This skill should be used when the user asks about "reading large files", "file too big", "context limit", "chunking files", "offset and limit", "file causing crash", or when encountering files over 10MB. Provides strategies for safely reading large files without exhausting context. |
| version | 1.0.0 |
Large File Handling for Claude Code
Context Limits
Claude Code has these file reading constraints:
- 25,000 tokens maximum per file read (~100KB of text)
- 200,000 tokens total context window
- Files over 50MB may cause system hangs
Safe Reading Strategies
Use Offset and Limit Parameters
Read specific portions instead of entire files:
Read file with offset=0 limit=500 # First 500 lines
Read file with offset=1000 limit=500 # Lines 1000-1500
Search Before Reading
Find relevant sections first:
grep -n "pattern" large-file.log | head -20
Then read only around matching line numbers.
Blocked File Types
Never attempt to read directly:
- Databases: .db, .sqlite, .sqlite3
- Archives: .zip, .tar, .gz, .7z
- Binaries: .exe, .dll, .so, .bin
- Large media: videos, large images
Alternative Approaches by File Type
| File Type | Alternative |
|---|---|
| SQLite DB | sqlite3 file.db "SELECT * FROM table LIMIT 10" |
| Large log | tail -1000 file.log or grep "ERROR" file.log |
| JSON (large) | jq '.specific.path' file.json |
| CSV (large) | head -100 file.csv or csvtool |
| PDF (large) | pdftotext file.pdf - | head -1000 |
| Minified JS | Usually not needed - read source instead |
.claudeignore Best Practices
Add these patterns to prevent accidental reads:
# Binary and database
*.db
*.sqlite
*.exe
*.dll
# Large generated files
*.min.js
*.bundle.js
# Dependencies
node_modules/
vendor/
.venv/
# Build output
dist/
build/
.next/
# Logs
*.log
logs/
Commands Available
/file-check <path>- Check if a file is safe to read/file-chunk <path>- Get chunking strategy for large file/scan-large-files [dir]- Find problematic files in directory
When Encountering Large Files
- Check size first using
/file-check - For text files > 10MB: Use chunking with offset/limit
- For logs: Search for relevant patterns first
- For databases: Use CLI tools to query
- For generated files: Often not needed - read source instead