.claude/skills/pr-demo/SKILL.md
Use when creating animated demos (GIFs) for pull requests or documentation. Covers terminal recording with asciinema and conversion to GIF/SVG for GitHub embedding.
npx skillsauth add mikeyobrien/ralph-orchestrator pr-demoInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Create polished terminal demos for PRs using asciinema recordings converted to GIF. The workflow: script → record → convert → embed.
| Goal | Tool Chain | Output | |------|------------|--------| | CLI demo for GitHub PR | asciinema → agg | GIF (< 5MB) | | Smaller file needed | asciinema → svg-term-cli | SVG (< 500KB) | | TUI screenshot | tmux → freeze | SVG/PNG |
Default choice: asciinema + agg (best compatibility, GitHub renders GIFs natively)
# Install tools (macOS)
brew install asciinema
cargo install --git https://github.com/asciinema/agg
npm install -g svg-term-cli # Optional: for SVG output
Before recording, write a brief script:
## Demo: [feature name]
Duration: ~20-30 seconds
1. [0-3s] Show command being typed
2. [3-10s] Command executes, show key output
3. [10-25s] Highlight the "aha moment" - what makes this valuable
4. [25-30s] Clean exit or final state
Keep it short. 20-30 seconds max. Show ONE thing well.
# Clean terminal state
clear
export PS1='$ ' # Simple prompt
export TERM=xterm-256color # Consistent colors
# Hide sensitive info (API keys, paths with usernames)
Terminal size: 100x24 (readable when scaled down)
# Record to .cast file
asciinema rec demo.cast --cols 100 --rows 24
# Execute your scripted demo
# Press Ctrl+D or type 'exit' when done
Tips:
# Basic conversion (recommended)
agg demo.cast demo.gif
# With speed adjustment (1.5x faster)
agg --speed 1.5 demo.cast demo.gif
# With custom font size for readability
agg --font-size 14 demo.cast demo.gif
Alternative - SVG (smaller files):
svg-term --in demo.cast --out demo.svg --window
Claude can self-validate demos using three approaches:
# Check file size (must be < 5MB for GitHub)
ls -lh demo.gif
# Check recording duration from .cast metadata
head -1 demo.cast | jq '.duration // "check manually"'
Extract a static frame for Claude to analyze:
# Option 1: Use svg-term to render a specific timestamp (e.g., 15 seconds in)
svg-term --in demo.cast --out demo-preview.svg --at 15000
# Option 2: Use asciinema cat + freeze for a snapshot
asciinema cat demo.cast | head -500 | freeze -o demo-preview.png
# Option 3: Just convert to GIF and use the file directly
# Claude can read GIF files with the Read tool
Then ask Claude to analyze using the Read tool on the image:
Validation prompt:
Analyze this terminal demo screenshot. Check:
1. Is the text readable (not too small/blurry)?
2. Is the command being demonstrated visible?
3. Is there any sensitive info (API keys, /Users/username paths)?
4. Does the terminal look clean (simple prompt, no clutter)?
5. Is the "aha moment" visible - what value does this demo show?
Rate: PASS or FAIL with specific issues.
The .cast file is JSON lines - validate the content programmatically:
# Check what commands were typed (input events)
grep '"i"' demo.cast | head -20
# Check recording duration
head -1 demo.cast | jq -r '.duration | floor'
# Should be 20-30 seconds
# Look for sensitive patterns
grep -iE '(api.?key|password|secret|/Users/[a-z])' demo.cast && echo "WARNING: Sensitive data found!"
After running the above, verify:
## Demo

*Shows: [one-sentence description of what the demo shows]*
Store demos in docs/demos/ or assets/ directory.
| Setting | Recommended Value | |---------|------------------| | Duration | 20-30 seconds | | Terminal size | 100x24 | | Speed multiplier | 1.0-1.5x | | Target file size | < 2MB ideal, < 5MB max | | Font size (agg) | 14-16 |
| Mistake | Fix | |---------|-----| | Demo too long | Script it first, show ONE thing | | Text unreadable | Use --font-size 14+, terminal 100x24 | | File too large | Use svg-term-cli instead, or increase speed | | Cluttered terminal | Clean PS1, clear history, hide paths | | No context in PR | Add one-line description below GIF |
docs/demos/
├── feature-name.gif # The demo
├── feature-name.cast # Source recording (optional, for re-rendering)
└── README.md # Recording instructions for future maintainers
development
Introspect, explain, and improve Ralph Orchestrator using its published llms.txt doc map. Use this skill whenever the user asks questions about Ralph's behavior, wants to understand how a Ralph internal works (event loop, hats, memories, tasks, backends, presets), debug an unfamiliar failure mode, or propose a code change to the ralph-orchestrator repo. The skill teaches the agent to discover authoritative answers from the live docs via llms.txt before guessing, and to scope improvements through the published architecture rather than the local checkout alone.
development
Run, monitor, resume, merge, and debug Ralph loops. Use this skill whenever the user asks to operate `ralph run` or `ralph loops`, inspect loop state, recover suspended loops, analyze diagnostics, or unblock merge queue issues.
development
Create, inspect, validate, explain, and improve Ralph hat collections. Use this skill whenever the user asks to make or refine a `.ralph/hats/*.yml` workflow, debug hat routing, explain event topology, or tune a multi-hat Ralph run.
testing
A directory-style test skill for smoke testing