Stats
Actions
Tags
From pell
Use when about to explain something inherently visual — architecture or data-flow diagrams, state machines, geometry or layout, before/after comparisons, tree or graph structures, timelines, or table-heavy comparisons — and a rendered view in a browser would communicate it better than terminal prose. Also use when soliciting a choice that's clearer as clickable options than as text.
How this skill is triggered — by the user, by Claude, or both
Slash command
/pell:visual-scratchpadThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Render the concept to a live browser pad instead of (or alongside) explaining it in text.
Render the concept to a live browser pad instead of (or alongside) explaining it in text.
Ensure the server is up. Check ~/.claude/pell-visual/server.pid; if it names no live process, start the bundled server in the background (Bash tool, run_in_background):
python3 "${CLAUDE_SKILL_DIR}/server.py" >/dev/null 2>&1
Then read ~/.claude/pell-visual/server.port for the bound port. If python3 is unavailable, skip the canvas and explain in the terminal — never block the task.
Write the fragment to ~/.claude/pell-visual/scratch.html (overwrite). Choose the format by leading character:
< → rendered as HTML (use for SVG/diagrams, precise layout, color, interactive buttons).Mention the URL once per session: http://127.0.0.1:<port>.
Start watch mode (default). If you have NOT already started the inbox watcher this session, start a zero-token shell watcher via the Monitor tool (persistent: true), streaming new lines so each click reacts in near-real-time:
tail -n0 -F ~/.claude/pell-visual/inbox.jsonl
Each line is a browser event {"ts":…, "payload":…} — parse the payload and react. At most one watcher per session — if one is already running, skip this step. Never poll with a recurring agent, loop, or cron.
To get a choice back, include interactive HTML that calls pellSend(...):
<button onclick="pellSend('variant-2')">Variant 2</button>
With watch mode on (step 4), clicks react in near-real-time. They also always surface on the user's next turn via the plugin's UserPromptSubmit hook, so nothing is lost even if the watcher isn't running.
tail -F streamed via Monitor; zero cost while idle, one watcher per session. The user can stop it with /pell:visualize stop-watch.npx claudepluginhub froncekb/pell_skills --plugin pellGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.