freeze-capture - SKILL.md Agent Skill

name: freeze-capture description: Capture terminal output or code as a static image (PNG/SVG/WebP) using charmbracelet/freeze. Use for proof-of-work screenshots when an agent finishes a task ("here's the green test run"), or for polished CLI/TUI demo stills for blog posts/social. NOT for animated demos — use terminal-demo-recorder (VHS) for those. Triggers on "freeze", "screenshot the output", "capture this command", "proof of work image", "make an image of", "snapshot the terminal".

Freeze Capture

Static image of terminal output or code file. Two modes, same tool.

Decide mode first

Goal	Mode	Output
Agent prove task worked (test passed, build succeeded, command ran)	proof	PNG, plain, no chrome
Demo for blog/social/README	demo	PNG/SVG, window chrome, theme, shadow
Code snippet from file (no execution)	code	SVG, syntax-highlighted

Proof-of-work (default for agents)

Goal: cheap, readable image showing command + output. Drop in task folder.

freeze --execute "<COMMAND>" \
  --output ".ai/$(date +%Y-%m-%d)/proof/<descriptive-name>.png" \
  --width 120 \
  --padding 20 \
  --background "#1e1e2e" \
  --theme "catppuccin-mocha"

Rules:

Path: .ai/YYYY-MM-DD/proof/ — create dir if missing.
Name: describe what's proven (pnpm-test-pass.png, nx-affected-build.png). No timestamps in name (date already in path).
Width 120 cols handles most CLI output without wrapping. Bump to 160 for wide tables.
PNG, not SVG — proof images get embedded in PR descriptions / pasted into Slack. PNG renders everywhere.
If command exits non-zero, still capture it. Failure proof is also proof.
Reference the file path back to the user so they can open it.

Example flow when agent wraps task:

mkdir -p .ai/$(date +%Y-%m-%d)/proof
freeze --execute "nx test my-lib" -o .ai/$(date +%Y-%m-%d)/proof/my-lib-test.png -W 140 -p 20

Demo (CLI/TUI showcase)

Goal: polished image for README, blog, X post. Window chrome on, shadow, generous padding.

freeze --execute "<COMMAND>" \
  --output demo.png \
  --window \
  --background "#0b0b1a" \
  --theme "dracula" \
  --padding 40,60 \
  --margin 30 \
  --border.radius 8 \
  --shadow.blur 20 --shadow.x 0 --shadow.y 10 \
  --font.family "JetBrains Mono" --font.size 16

Tune:

Bigger font (16-20) for social where image gets downsized.
--width and --height to enforce 16:9 or square aspect for X/LinkedIn.
Strip ANSI cursor codes if TUI leaves artifacts: pipe through sed -r "s/\x1B\[[0-9;]*[a-zA-Z]//g" before capture, or use script for full PTY.

Code-from-file

freeze src/lib/foo.ts --output foo.svg --window --show-line-numbers --theme dracula
freeze src/lib/foo.ts --output foo.png --lines 12,40   # only lines 12-40

SVG preferred for code — scales crisp, embeds in docs.

Caveats

TUI capture is finicky. Programs that redraw (htop, vim, nx graph) won't render their final state via --execute. For TUIs use VHS (terminal-demo-recorder skill) instead.
PNG file size: a 120-col x ~30-row capture is ~5-8 MB at default settings. Drop --font.size to 12 or use SVG to shrink.
Color: freeze respects ANSI in --execute output. If a CLI suppresses color when stdout isn't a TTY, force it (FORCE_COLOR=1, --color always, etc.) before piping into --execute.
No animation. Single frame only. If user asks for "a recording" or "GIF", redirect to terminal-demo-recorder.
Exit code is preserved by --execute — image still produced.

After capture

Print the absolute path of the output file so the user can open it.
Bump ~/projects/dot-ai-config/USAGE.md row for freeze-capture (date + count).

More flags

See references/flags.md for the full list and references/recipes.md for copy-paste presets.