mikeyobrien

tui-validate

@mikeyobrien/tui-validate
mikeyobrien
815
109 forks
Updated 1/18/2026
View on GitHub

Validates Terminal User Interface (TUI) output using freeze for screenshot capture and LLM-as-judge for semantic validation. Supports both visual (PNG/SVG) and text-based validation modes.

Installation

$skills install @mikeyobrien/tui-validate
Claude Code
Cursor
Copilot
Codex
Antigravity

Details

Path.claude/skills/tui-validate/SKILL.md
Branchmain
Scoped Name@mikeyobrien/tui-validate

Usage

After installing, this skill will be available to your AI coding assistant.

Verify installation:

skills list

Skill Instructions


name: tui-validate description: Validates Terminal User Interface (TUI) output using freeze for screenshot capture and LLM-as-judge for semantic validation. Supports both visual (PNG/SVG) and text-based validation modes. type: anthropic-skill version: "1.0"

TUI Validate

Overview

This skill validates Terminal User Interface (TUI) applications by capturing their output and using LLM-as-judge for semantic validation. It leverages freeze from Charmbracelet for high-fidelity terminal screenshots and provides structured validation criteria.

Philosophy: Rather than brittle string matching, this skill uses semantic understanding to validate that TUI output "looks right" - checking layout, content presence, and visual hierarchy without breaking on minor formatting changes.

When to Use

  • Validating TUI rendering after changes
  • Checking that UI components display correctly
  • Visual regression testing for terminal applications
  • Verifying TUI state after specific interactions
  • Creating documentation screenshots with validation

Prerequisites

Required:

  • freeze CLI tool installed (brew install charmbracelet/tap/freeze)
  • tmux for interactive TUI capture (optional, for live applications)

Verification:

# Check freeze is installed
freeze --version

# Check tmux is installed (for interactive capture)
tmux -V

Parameters

  • target (required): What to validate. One of:

    • file:<path> - ANSI output file to validate
    • command:<cmd> - Command to execute and capture
    • tmux:<session> - Live tmux session to capture
    • buffer:<text> - Raw text/ANSI to validate
  • criteria (required): Validation criteria. Can be:

    • A predefined criteria name (see Built-in Criteria)
    • A custom criteria string describing what to check
  • output_format (optional, default: "svg"): Screenshot format

    • svg - Vector format, best for documentation
    • png - Raster format, best for visual diff
    • text - Text-only extraction, fastest
  • save_screenshot (optional, default: false): Whether to save the screenshot

    • If true, saves to {target_name}.{format} in current directory
  • judge_mode (optional, default: "semantic"): Validation approach

    • semantic - LLM judges based on meaning and layout
    • strict - Also checks exact content presence
    • visual - Requires PNG, checks visual appearance

Built-in Criteria

ralph-header

Validates Ralph TUI header component:

  • Iteration counter in [iter N] or [iter N/M] format
  • Elapsed time in MM:SS format
  • Hat indicator with emoji and name
  • Mode indicator (▶ auto or ⏸ paused)
  • Optional scroll mode indicator [SCROLL]
  • Optional idle countdown idle: Ns

ralph-footer

Validates Ralph TUI footer component:

  • Activity indicator (◉ active, ◯ idle, or ■ done)
  • Last event topic display
  • Search mode display when active

ralph-full

Validates complete Ralph TUI layout:

  • Header section at top (3 lines)
  • Terminal content area (variable height)
  • Footer section at bottom (3 lines)
  • Proper visual hierarchy and borders

tui-basic

Generic TUI validation:

  • Has visible content (not blank)
  • No rendering artifacts or broken characters
  • Proper terminal dimensions

Execution Flow

1. Capture Phase

Capture TUI output based on target type:

For file targets:

freeze {file_path} -o /tmp/tui-capture.{format}

For command targets:

freeze --execute "{command}" -o /tmp/tui-capture.{format}

For tmux targets:

tmux capture-pane -pet {session} | freeze -o /tmp/tui-capture.{format}

For buffer targets:

echo "{buffer}" | freeze -o /tmp/tui-capture.{format}

Constraints:

  • You MUST verify freeze is installed before attempting capture
  • You MUST handle capture failures gracefully and report the error
  • You MUST use appropriate freeze flags for the output format
  • You SHOULD use --theme base16 for consistent rendering
  • You SHOULD set reasonable dimensions with --width and --height

2. Extraction Phase

Extract content for LLM analysis:

For text/semantic validation:

  • If format is text, use the captured text directly
  • If format is svg or png, also capture text version for content analysis

For visual validation:

  • Requires PNG format
  • Will analyze the image directly using vision capabilities

Constraints:

  • You MUST extract both visual and text representations when judge_mode is visual
  • You MUST preserve ANSI escape sequences for color validation when relevant

3. Validation Phase

Apply LLM-as-judge with the appropriate criteria:

Semantic Validation Prompt Template:

Analyze this terminal UI output and determine if it meets the following criteria:

CRITERIA:
{criteria_description}

TERMINAL OUTPUT:
{captured_text}

Evaluate each criterion and provide:
1. PASS or FAIL for each requirement
2. Brief explanation for any failures
3. Overall verdict: PASS or FAIL

Be lenient on exact formatting but strict on:
- Required content presence
- Logical layout and hierarchy
- No rendering errors or artifacts

Visual Validation Prompt Template (with image):

Examine this terminal screenshot and validate:

CRITERIA:
{criteria_description}

Check for:
1. Visual hierarchy and layout
2. Color coding correctness
3. No rendering artifacts or broken characters
4. Proper alignment and spacing

Verdict: PASS or FAIL with explanation

Constraints:

  • You MUST return a clear PASS or FAIL verdict
  • You MUST provide specific feedback on failures
  • You MUST be lenient on whitespace/formatting differences
  • You MUST be strict on content presence and semantic correctness
  • You SHOULD note any warnings even on PASS results

4. Reporting Phase

Report validation results:

On PASS:

✅ TUI Validation PASSED

Criteria: {criteria_name}
Target: {target}
Mode: {judge_mode}

All requirements satisfied.
{optional_notes}

On FAIL:

❌ TUI Validation FAILED

Criteria: {criteria_name}
Target: {target}
Mode: {judge_mode}

Issues found:
- {issue_1}
- {issue_2}

Screenshot saved: {path_if_saved}

Constraints:

  • You MUST always provide a clear verdict
  • You MUST list specific issues on failure
  • You MUST offer the screenshot path if saved
  • You SHOULD suggest fixes for common issues

Examples

Example 1: Validate Ralph Header from File

Input:

/tui-validate file:test_output.txt criteria:ralph-header

Process:

  1. Read test_output.txt containing ANSI output
  2. Capture with freeze: freeze test_output.txt -o /tmp/capture.svg
  3. Extract text content
  4. Apply ralph-header criteria via LLM judge
  5. Report PASS/FAIL with details

Example 2: Validate Live TUI in tmux

Input:

/tui-validate tmux:ralph-session criteria:ralph-full save_screenshot:true

Process:

  1. Capture tmux pane: tmux capture-pane -pet ralph-session | freeze -o ralph-session.svg
  2. Also capture text: tmux capture-pane -pet ralph-session > /tmp/text.txt
  3. Apply ralph-full criteria checking header, content, and footer
  4. Save screenshot to ralph-session.svg
  5. Report validation result

Example 3: Custom Criteria Validation

Input:

/tui-validate command:"cargo run --example tui_demo" criteria:"Shows a bordered box with 'Hello World' text centered inside" output_format:png judge_mode:visual

Process:

  1. Execute command and capture: freeze --execute "cargo run --example tui_demo" -o /tmp/capture.png
  2. Use vision model to analyze PNG
  3. Check for bordered box and centered text
  4. Report visual validation result

Example 4: Quick Text Validation

Input:

/tui-validate buffer:"[iter 3/10] 04:32 | 🔨 Builder | ▶ auto" criteria:ralph-header output_format:text

Process:

  1. Analyze text directly (no freeze needed for text mode with buffer)
  2. Check for iteration format, elapsed time, hat, and mode indicator
  3. Report validation result

Criteria Definitions

ralph-header (Full Definition)

name: ralph-header
description: Ralph TUI header component validation
requirements:
  - name: iteration_counter
    description: Shows iteration in [iter N] or [iter N/M] format
    required: true
    pattern: '\[iter \d+(/\d+)?\]'

  - name: elapsed_time
    description: Shows elapsed time in MM:SS format
    required: true
    pattern: '\d{2}:\d{2}'

  - name: hat_indicator
    description: Shows current hat with emoji prefix
    required: true
    examples: ["🔨 Builder", "📋 Planner", "🎯 Executor"]

  - name: mode_indicator
    description: Shows loop mode status
    required: true
    values: ["▶ auto", "⏸ paused"]

  - name: scroll_indicator
    description: Shows [SCROLL] when in scroll mode
    required: false
    pattern: '\[SCROLL\]'

  - name: idle_countdown
    description: Shows idle timeout when present
    required: false
    pattern: 'idle: \d+s'

ralph-footer (Full Definition)

name: ralph-footer
description: Ralph TUI footer component validation
requirements:
  - name: activity_indicator
    description: Shows current activity state
    required: true
    values: ["◉ active", "◯ idle", "■ done"]

  - name: event_topic
    description: Shows last event topic
    required: false
    examples: ["task.start", "build.done", "loop.terminate"]

  - name: search_display
    description: Shows search query and match count when searching
    required: false
    pattern: 'Search: .+ \d+/\d+'

ralph-full (Full Definition)

name: ralph-full
description: Complete Ralph TUI layout validation
requirements:
  - name: header_section
    description: Header at top with iteration, time, hat, and mode
    required: true
    references: ralph-header

  - name: content_section
    description: Main terminal content area
    required: true
    checks:
      - Has visible content or is ready for content
      - Properly bounded between header and footer

  - name: footer_section
    description: Footer at bottom with activity status
    required: true
    references: ralph-footer

  - name: visual_hierarchy
    description: Clear visual separation between sections
    required: true
    checks:
      - Borders or spacing between sections
      - Consistent width across sections

Troubleshooting

freeze not found

# macOS
brew install charmbracelet/tap/freeze

# Linux (via Go)
go install github.com/charmbracelet/freeze@latest

# Verify installation
freeze --version

tmux capture fails

  • Ensure the tmux session exists: tmux list-sessions
  • Verify pane number: tmux list-panes -t {session}
  • Try capturing specific pane: tmux capture-pane -pet {session}:{pane}

Rendering artifacts in capture

  • Try different terminal emulator settings in freeze
  • Use --theme flag for consistent colors
  • Ensure terminal dimensions match TUI expectations

LLM judge too strict/lenient

  • Adjust criteria to be more specific
  • Use strict mode for exact matching requirements
  • Use semantic mode for layout/presence checking

Integration with Tests

This skill can be integrated into test suites:

// In tests/tui_validation.rs
#[test]
#[ignore] // Run with: cargo test -- --ignored
fn validate_header_rendering() {
    // 1. Render header to buffer
    let output = render_header_to_string(&test_state);

    // 2. Save to temp file
    std::fs::write("/tmp/header_test.txt", &output).unwrap();

    // 3. Run tui-validate skill (via CLI or programmatic)
    // /tui-validate file:/tmp/header_test.txt criteria:ralph-header

    // 4. Assert validation passed
}

Best Practices

  1. Use semantic validation for layout checks - Don't break on minor formatting
  2. Use strict validation for content requirements - Ensure critical info is present
  3. Save screenshots on failure - Aids debugging
  4. Test with various terminal sizes - TUIs should be responsive
  5. Combine with unit tests - Use this for integration/visual validation, unit tests for logic