textual-capture 0.3.0

Sequenced screen capture for Textual TUI applications. Enables AI assistants to review and test TUIs.

textual-capture

Sequenced screenshot capture for Textual TUI applications

textual-capture automates UI interactions in your Textual apps and captures screenshots at key moments. Define sequences of key presses, clicks, and delays in simple TOML files.

Perfect for:

• 🤖 LLM-driven TUI review and testing (AI can generate configs and analyze output)
• 📸 Documentation screenshots (consistent, reproducible captures)
• 🎬 Demo creation (step-through your app automatically)
• ✅ Visual regression prep (capture baseline states)

---

Quick Start

Installation

pip install textual-capture

Your First Capture

Create demo.toml:

app_module = "my_app"
app_class = "MyApp"

[[step]]
type = "press"
keys = ["tab", "tab", "enter"]

[[step]]
type = "capture"
output = "my_screenshot"

Run it:

textual-capture demo.toml

This creates:

• my_screenshot.svg (visual)
• my_screenshot.txt (text representation)
• my_screenshot_tooltips.txt (widget tooltips)

---

Core Features

Multi-Step Sequences

Chain actions together:

[[step]]
type = "press"
keys = ["ctrl+n"]         # Open new dialog

[[step]]
type = "delay"
seconds = 0.5             # Wait for animation

[[step]]
type = "click"
label = "Submit"          # Click button

[[step]]
type = "capture"          # Take screenshot

Auto-Sequencing

Omit output names for automatic numbering:

[[step]]
type = "capture"
# Creates: capture_001.svg, capture_001.txt, capture_001_tooltips.txt

[[step]]
type = "press"
keys = ["down"]

[[step]]
type = "capture"
# Creates: capture_002.svg, capture_002.txt, capture_002_tooltips.txt

Keyboard Shortcuts

Full modifier support:

[[step]]
type = "press"
keys = ["ctrl+s", "ctrl+shift+p", "alt+f4"]

Smart Tooltips

Tooltips captured automatically with every screenshot (opt-out if not needed):

# Enabled by default - just works!
[[step]]
type = "capture"

# Disable if not needed
[[step]]
type = "capture"
capture_tooltips = false

# Capture only tooltips (fast!)
[[step]]
type = "capture"
formats = []              # Skip SVG/TXT
capture_tooltips = true

Organized Output

output_dir = "./screenshots"   # All files go here
formats = ["svg"]              # Only generate SVG (faster)

Validation

Check your config before running:

textual-capture demo.toml --dry-run

Shows planned steps, validates imports, catches errors.

---

Common Use Cases

Documentation Screenshots

output_dir = "./docs/screenshots"
formats = ["svg"]

[[step]]
type = "capture"
output = "main_menu"

[[step]]
type = "press"
keys = ["tab", "enter"]

[[step]]
type = "capture"
output = "settings_dialog"

LLM UI Analysis

output_dir = "./llm_review"
formats = ["txt"]           # Text for AI analysis
capture_tooltips = true     # Include tooltip data

[[step]]
type = "capture"
# AI can read the text files and tooltips

Then: cat llm_review/*.txt | claude analyze-ui

Accessibility Audit

formats = []                     # Skip visuals
capture_tooltips = true
tooltip_include_empty = true     # Show missing tooltips

[[step]]
type = "capture"
output = "tooltip_audit"

Review tooltip_audit_tooltips.txt for widgets with (no tooltip).

Keyboard Workflow Testing

[[step]]
type = "press"
keys = ["ctrl+o"]           # Open file

[[step]]
type = "delay"
seconds = 0.5

[[step]]
type = "press"
keys = ["t", "e", "s", "t"]
pause_between = 0.1           # Slow typing

[[step]]
type = "press"
keys = ["enter"]

[[step]]
type = "capture"
output = "file_opened"

---

Configuration Reference

Required Fields

app_module = "path.to.module"      # Python module with your app
app_class = "MyApp"                # Textual App class name

Global Settings

# Screen size
screen_width = 100                 # Default: 80
screen_height = 40                 # Default: 40

# Timing
initial_delay = 1.0                # Wait before first action (default: 1.0)

# Behavior
scroll_to_top = true               # Press "home" at start (default: true)
module_path = "."                  # Add to sys.path (optional)

# Output
output_dir = "./screenshots"       # Where to save files (default: ".")
formats = ["svg", "txt"]           # Formats to generate (default: both)

# Tooltips
capture_tooltips = true            # Capture tooltips (default: true)
widget_selector = "*"             # CSS selector (default: all widgets)
tooltip_include_empty = false      # Show widgets without tooltips (default: false)

Action Types

Press Keys

[[step]]
type = "press"
keys = ["tab", "ctrl+s", "enter"]  # List syntax (preferred)
pause_between = 0.2                  # Seconds between keys (default: 0.2)

# Legacy comma-separated syntax still works
key = "tab,ctrl+s,enter"

Supported modifiers: ctrl+, shift+, alt+, meta+

Click Button

[[step]]
type = "click"
label = "Submit"                   # Button text (spaces removed for ID)

Delay

[[step]]
type = "delay"
seconds = 1.5                      # Seconds to wait

Capture Screenshot

[[step]]
type = "capture"
output = "my_state"                # Optional: custom name
formats = ["svg", "txt"]           # Optional: override global
capture_tooltips = true            # Optional: override global
widget_selector = "Button"        # Optional: custom selector
tooltip_include_empty = false      # Optional: override global

If output is omitted, auto-generates capture_001, capture_002, etc.

---

Advanced Features

Selective Formats

Generate only what you need:

# Global default
formats = ["svg"]  # Only SVG (faster)

[[step]]
type = "capture"
output = "visual_only"
# Uses global: svg only

[[step]]
type = "capture"
output = "complete"
formats = ["svg", "txt"]  # Override: both formats

Valid formats: svg, txt

Tooltip Configuration

Fine-tune tooltip capture:

# Capture all widgets with tooltips
[[step]]
type = "capture"
output = "all_tooltips"

# Capture only buttons
[[step]]
type = "capture"
output = "button_tooltips"
widget_selector = "Button"

# Include widgets without tooltips
[[step]]
type = "capture"
output = "complete_audit"
tooltip_include_empty = true

Tooltip file format:

# Tooltips captured from: my_state
# Selector: *
# Timestamp: 2025-12-20 10:30:45

Button#run: Start the selected command
Button#cancel: Abort operation (Esc)
Input#search: Search for items
Label#status: (no tooltip)

Tooltip-Only Captures

Skip expensive rendering for fast metadata extraction:

formats = []              # No SVG or TXT
capture_tooltips = true

[[step]]
type = "capture"
output = "metadata_only"
# Creates only: metadata_only_tooltips.txt

Use cases:

• Fast UI audits
• Tooltip validation
• LLM analysis pipelines
• Documentation generation

---

CLI Usage

# Run capture sequence
textual-capture config.toml

# Show all actions as they execute
textual-capture config.toml --verbose

# Suppress all output except errors
textual-capture config.toml --quiet

# Validate config without running
textual-capture config.toml --dry-run

Dry-run output example:

Configuration: demo.toml
App: my_app.MyApp
Screen: 80x40
Output Directory: ./screenshots
Default Formats: svg, txt
Capture Tooltips: True
Tooltip Selector: *

Planned Steps (4 total):
  1. press: keys=['tab', 'tab', 'enter']
  2. delay: 0.5s
  3. capture: output="my_state", formats=[svg, txt], tooltips=*
  4. press: keys=['ctrl+q']

Validating module import...
✓ Successfully imported MyApp from my_app

✓ Configuration valid and ready to execute

---

🤖 LLM-Driven Workflows

Primary Use Case: AI assistants can generate configs, run captures, and analyze output.

How LLMs Use textual-capture

1. Generate TOML config based on user request
2. Run with --dry-run to validate
3. Execute capture to get screenshots + tooltips
4. Analyze text files (.txt and _tooltips.txt)
5. Report findings to user

Example: Claude Code Integration

User: "Check if my settings dialog has proper tooltips"

Claude:

1. Creates settings_check.toml:

app_module = "your_app"
app_class = "YourApp"
formats = []
capture_tooltips = true
tooltip_include_empty = true

[[step]]
type = "press"
keys = ["ctrl+comma"]  # Open settings

[[step]]
type = "capture"
output = "settings_tooltips"

2. Runs: textual-capture settings_check.toml
3. Reads: settings_tooltips_tooltips.txt
4. Reports: "Found 3 buttons without tooltips: Button#apply, Button#reset, Button#advanced"

Pro tip: Copy LLM_INSTRUCTIONS.md into your project's CLAUDE.md file to give Claude Code full context.

---

Comparison with Other Tools

Feature	textual-dev screenshot	pytest-textual-snapshot	textual-capture
Single capture	Yes	Yes	Yes
Multi-step sequences	No	No	✅ Yes
Keyboard shortcuts	No	No	✅ Yes
Button clicks	No	No	✅ Yes
Delays/timing	No	No	✅ Yes
Auto-sequencing	No	No	✅ Yes
Tooltip capture	No	No	✅ Yes
Dry-run validation	No	No	✅ Yes
Organized output	No	No	✅ Yes
LLM-friendly	No	No	✅ Yes
Human-readable config	No	No	✅ Yes

---

Tips & Best Practices

For Documentation

• Use formats = ["svg"] for faster generation
• Use descriptive output names: output = "main_menu"
• Keep sequences focused on one feature/flow

For LLM Analysis

• Use formats = ["txt"] + capture_tooltips = true
• Use tooltip_include_empty = true for audits
• Auto-sequence unnamed captures for exploration

For Testing

• Use --dry-run during development
• Use output_dir to keep project root clean
• Validate configs in CI with dry-run

For Performance

• Use selective formats: formats = ["svg"] or ["txt"]
• Use formats = [] for tooltip-only captures
• Reduce initial_delay if your app renders quickly

---

Examples

See the examples/ directory for:

• llm_review.toml - LLM-driven UI analysis template
• keyboard_shortcuts.toml - Complex keyboard workflows
• tooltip_audit.toml - Accessibility checking
• documentation.toml - Multi-capture documentation

---

Contributing

Contributions welcome! See CONTRIBUTING.md (coming soon).

Issues and feature requests: https://github.com/eyecantell/textual-capture/issues

---

License

MIT © 2025 Paul Neumann