pyrex41/claude_skills/visualization

Ralph Visualization

Add a stream display TUI to monitor Ralph loop iterations. This is an optional layer on top of the core Ralph setup.

Prerequisites

• Existing Ralph loop setup (run /ralph-setup first)
• Claude Code harness (stream display requires --output-format stream-json)

What This Adds

1. `stream_display.py` - TUI that renders streaming JSON as readable output
2. Updates loop.sh to pipe through the display (if not already configured)

Features

• Tool call display - Color-coded tool names with key arguments
• Text toggle - Press [v] during streaming to show/hide assistant text
• Git diff summary - Shows files changed at end of iteration
• Iteration stats - Duration, tool count, changes

Usage

After installation:

# Runs with display automatically
./loop.sh --build
 
# Disable display for raw output
./loop.sh --build --no-display
 
# Debug: dump raw JSON for inspection
./loop.sh --build --dump /tmp/stream.jsonl

Installation

When invoked, this skill will:

1. Check that Ralph is already set up in the target directory
2. Copy stream_display.py to the project
3. Verify loop.sh has display piping configured

Reference

See references/stream-display.md for architecture details.

Execution Steps

When this skill is invoked:

1. Check for existing Ralph setup:

``bash ls ralph.conf PROMPT_build.md loop.sh 2>/dev/null ` If not set up, tell user to run /ralph-setup` first.

1. Copy stream_display.py to project:

Read from this skill's directory and write to project root:

• Source: skills/visualization/stream_display.py
• Target: $PROJECT/stream_display.py

1. Update loop.sh to enable stream display:

Add these variables near the top (after STOP_FILE): ``bash # Stream display (optional - requires Claude Code harness) DISPLAY_SCRIPT="${DISPLAY_SCRIPT:-stream_display.py}" DUMP_FILE="${DUMP_FILE:-}" ``

Update the claude command in build_command() to include streaming flags: ``bash local cmd="cat '$prompt_file' | claude --output-format stream-json --verbose" ``

Update the main loop's run section to pipe through display: ``bash if [[ -n "$DISPLAY_SCRIPT" && -f "$DISPLAY_SCRIPT" && "$HARNESS" == "claude" ]]; then DISPLAY_ARGS="--iteration $iteration --mode $MODE" [[ -n "$DUMP_FILE" ]] && DISPLAY_ARGS="$DISPLAY_ARGS --dump $DUMP_FILE" eval "$cmd" | python3 "$DISPLAY_SCRIPT" $DISPLAY_ARGS else eval "$cmd" fi ``

1. Add CLI flags to usage and argument parsing:

• --dump <file> - Dump raw JSON for debugging
• --no-display - Disable stream display

1. Show usage instructions:

• Press [v] during streaming to toggle text visibility
• Use --dump /tmp/stream.jsonl to debug parsing issues
• Only works with Claude Code harness (--output-format stream-json)