v1.3.0 - Sharper triggering and readable nodes

Two changes, both from user feedback.

Triggering is tighter. The skill used to fire on any broad ask for ideas or brainstorming, which caught people who actually wanted concrete answers. Now it fires when you signal you want to open a space and think alongside it, and it holds back when the request names a deliverable, a count, a deadline, or asks which option to pick. When intent is genuinely unclear it asks one line first instead of branching. Calling it by name with /wonder or /wonderpill always runs it.

Mind-map nodes are readable on their own. Nodes were too terse to follow without clicking each one. Every node now carries a plain-language handle plus a short line saying what to push on, so the map reads on a first pass. The click is just for the deeper gut-check now.

Same method underneath: assumption audit, the inversion rule, tendrils, uneven depth, feral vs tethered, and the plausibility vs generativity scrap judgment. The universal variant and the README were updated to match.

v1.2.0 - Universal (platform-agnostic) skill

What's new

A single portable universal/SKILL.md — the same Wonder Pill method, but with every Claude-specific tool call stripped out so it works anywhere that takes written instructions: ChatGPT, Kimi, opencode, Antigravity, Z.ai, or any other assistant.

What changed to make it portable:

• The tappable intake UI (ask_user_input_v0) becomes three plain numbered questions asked in chat.
• The Claude Visualizer dependency is gone. Instead, the skill always builds one self-contained HTML+CSS+JS mind map file (no CDNs, no external calls) — shown inline if the host has a canvas/artifact pane, or just handed over as a file to open in a browser if it doesn't. Same file either way, different delivery path.
• A plain markdown outline is kept as a last-resort fallback for hosts that can't take a large HTML block at all.

What stayed exactly the same:

• The assumption audit, the inversion rule, tendrils across scale/time/reversal/audience/material/sense/causality, uneven depth, feral-vs-tethered tagging, and the plausibility-vs-generativity scrap judgment (derivation / flaw / judgment) from v1.1.0.

The original Claude-native SKILL.md and references/mindmap.md are untouched — this is an addition (universal/SKILL.md), not a replacement.

To use it: copy the contents of universal/SKILL.md into your platform's custom instructions / system prompt / agent config.

v1.1.0 - Honest scrap judgment

What changed

Someone pointed out a gap: the skill kills off "scrapped" what-ifs using the same model, in the same breath as generating them, with no outside check. That's exactly where a bias toward safe, plausible ideas over weird-but-valid ones could sneak in unnoticed.

Fixes in this release:

• Plausibility and generativity are now separate axes. An idea can only be scrapped for being a genuine dead end (nothing left to ask once you're standing in it) — never for just sounding unlikely or hard to build. Implausible-but-generative survives; plausible-but-flat can still die.
• Every scrapped what-if now shows its full reasoning, not one blended line: where it came from, what's actually wrong with it, and a separate statement of why that flaw was judged disqualifying.
• All of that now renders on the mind map, next to the struck-through idea, so you can see the model's reasoning and disagree with it if a kill call looks wrong.

Net effect: fewer silently-discarded weird ideas, and a paper trail when one does get cut.