File Watching
SwissArmyHammer includes a powerful file watching system that automatically detects and reloads prompt changes without restarting the server.
How It Works
The file watcher monitors your prompt directories for changes and automatically:
- Detects new, modified, or deleted prompt files
- Validates changed files for syntax errors
- Reloads prompts into memory
- Notifies connected clients of updates
- Maintains state during the reload process
┌─────────────┐ ┌──────────────┐ ┌──────────────┐
│ File System │────>│ File Watcher │────>│ Prompt Cache │
└─────────────┘ └──────────────┘ └──────────────┘
│ │
▼ ▼
┌──────────────┐ ┌──────────────┐
│ Validator │ │ MCP Clients │
└──────────────┘ └──────────────┘
Configuration
Basic Settings
Configure file watching in your config.yaml:
watch:
# Enable/disable file watching
enabled: true
# Check interval in milliseconds
interval: 1000
# Debounce delay to batch rapid changes
debounce: 500
# Maximum files to process per cycle
batch_size: 100
Advanced Options
watch:
# File patterns to watch
patterns:
- "**/*.md"
- "**/*.markdown"
- "**/prompts.yaml"
# Patterns to ignore
ignore:
- "**/node_modules/**"
- "**/.git/**"
- "**/target/**"
- "**/*.swp"
- "**/*~"
- "**/.DS_Store"
# Watch strategy
strategy: efficient # efficient, aggressive, polling
# Platform-specific settings
platform:
# macOS FSEvents
macos:
use_fsevents: true
latency: 0.1
# Linux inotify
linux:
use_inotify: true
max_watches: 8192
# Windows
windows:
use_polling: false
poll_interval: 1000
Watch Strategies
Efficient (Default)
Best for most use cases:
watch:
strategy: efficient
# Uses native OS file watching APIs
# Low CPU usage
# May have slight delay on some systems
Aggressive
For development with frequent changes:
watch:
strategy: aggressive
interval: 100 # Check every 100ms
debounce: 50 # Minimal debounce
# Higher CPU usage
# Near-instant updates
Polling
Fallback for compatibility:
watch:
strategy: polling
interval: 2000 # Poll every 2 seconds
# Works everywhere
# Higher CPU usage
# Slower updates
File Events
Supported Events
The watcher handles these file system events:
- Created - New prompt files added
- Modified - Existing files changed
- Deleted - Files removed
- Renamed - Files moved or renamed
- Metadata - Permission or timestamp changes
Event Processing
# Event processing configuration
watch:
events:
# Process creation events
create:
enabled: true
validate: true
# Process modification events
modify:
enabled: true
validate: true
reload_delay: 100 # ms
# Process deletion events
delete:
enabled: true
cleanup_cache: true
# Process rename events
rename:
enabled: true
track_moves: true
Validation
Automatic Validation
Files are validated before reload:
watch:
validation:
# Enable validation
enabled: true
# Validation rules
rules:
# Check YAML front matter
yaml_syntax: true
# Validate required fields
required_fields:
- name
- title
- description
# Check template syntax
template_syntax: true
# Maximum file size
max_size: 1MB
# What to do on validation failure
on_failure: warn # warn, ignore, stop
Validation Errors
When validation fails:
[WARN] Validation failed for prompts/invalid.md:
- Line 5: Invalid YAML syntax
- Missing required field: 'title'
- Template error: Unclosed tag '{% if'
File will not be loaded. Fix errors and save again.
Performance
Optimization Tips
-
Exclude unnecessary paths:
watch: ignore: - "**/backup/**" - "**/archive/**" - "**/*.log" -
Tune intervals for your workflow:
# For active development watch: interval: 500 debounce: 250 # For production watch: interval: 5000 debounce: 2000 -
Limit watch scope:
watch: # Only watch specific directories directories: - ./prompts - ~/.swissarmyhammer/prompts # Don't watch subdirectories recursive: false
Resource Usage
Monitor watcher resource usage:
# Check watcher status
swissarmyhammer doctor --watch
# Show watcher statistics
swissarmyhammer status --verbose
# Output:
File Watcher Status:
Strategy: efficient
Files watched: 156
Directories: 12
CPU usage: 0.1%
Memory: 2.4MB
Events processed: 1,234
Last reload: 2 minutes ago
Debugging
Enable Debug Logging
logging:
modules:
swissarmyhammer::watcher: debug
Common Issues
-
Changes not detected:
# Check if watching is enabled swissarmyhammer config get watch.enabled # Test file watching swissarmyhammer test --watch -
High CPU usage:
# Increase intervals watch: interval: 2000 debounce: 1000 # Use efficient strategy watch: strategy: efficient -
Too many open files:
# Linux: Increase inotify watches echo fs.inotify.max_user_watches=524288 | sudo tee -a /etc/sysctl.conf sudo sysctl -p # macOS: Usually not an issue with FSEvents # Windows: Use polling fallback
Platform-Specific Notes
macOS
Uses FSEvents for efficient watching:
watch:
platform:
macos:
use_fsevents: true
# FSEvents latency in seconds
latency: 0.1
# Ignore events older than
ignore_older_than: 10 # seconds
Linux
Uses inotify with automatic limits:
watch:
platform:
linux:
use_inotify: true
# Will warn if approaching limits
warn_threshold: 0.8
# Fallback to polling if needed
auto_fallback: true
Windows
Uses ReadDirectoryChangesW:
watch:
platform:
windows:
# Buffer size for changes
buffer_size: 65536
# Watch subtree
watch_subtree: true
# Notification filters
filters:
- file_name
- last_write
- size
Integration
Client Notifications
Clients are notified of changes:
// MCP client receives notification
client.on('prompt.changed', (event) => {
console.log(`Prompt ${event.name} was ${event.type}`);
// Refresh UI, clear caches, etc.
});
Hooks
Run commands on file changes:
watch:
hooks:
# Before processing changes
pre_reload:
- echo "Reloading prompts..."
# After successful reload
post_reload:
- ./scripts/notify-team.sh
- ./scripts/update-index.sh
# On reload failure
on_error:
- ./scripts/alert-admin.sh
API Access
Query watcher status via API:
# Get watcher status
curl http://localhost:3333/api/watcher/status
# Get recent events
curl http://localhost:3333/api/watcher/events
# Trigger manual reload
curl -X POST http://localhost:3333/api/watcher/reload
Best Practices
Development
- Use aggressive watching for immediate feedback
- Enable validation to catch errors early
- Watch only active directories to reduce overhead
- Use debug logging to troubleshoot issues
Production
- Use efficient strategy for lower resource usage
- Increase intervals to reduce CPU load
- Disable watching if prompts rarely change
- Monitor resource usage regularly
Large Projects
- Exclude build directories and dependencies
- Use specific patterns instead of wildcards
- Consider splitting prompts across multiple directories
- Implement caching to reduce reload impact
Manual Control
CLI Commands
Control file watching manually:
# Pause file watching
swissarmyhammer watch pause
# Resume file watching
swissarmyhammer watch resume
# Force reload all prompts
swissarmyhammer watch reload
# Show watch status
swissarmyhammer watch status
Environment Variables
Override watch settings:
# Disable watching
export SWISSARMYHAMMER_WATCH_ENABLED=false
# Change interval
export SWISSARMYHAMMER_WATCH_INTERVAL=5000
# Force polling strategy
export SWISSARMYHAMMER_WATCH_STRATEGY=polling
Troubleshooting
Diagnostic Commands
# Run watcher diagnostics
swissarmyhammer doctor --watch
# Test file detection
echo "test" >> prompts/test.md
swissarmyhammer watch test
# Monitor events in real-time
swissarmyhammer watch monitor
Common Solutions
- Linux: Increase inotify limits
- macOS: Grant full disk access
- Windows: Run as administrator
- All: Check file permissions
- All: Verify ignore patterns
Next Steps
- Configure watching in Configuration
- Learn about Prompt Organization
- Understand Prompt Overrides
- Read Troubleshooting for more help