spike

spike — measure the blackbox before you describe it

gflow-cli drives a product it does not own. Every statement about what Flow does is either measured or guessed, and a guess written into code or docs becomes a fact nobody re-checks.

Load this skill before you write, say, or encode any of these:

• "X is not supported / not available / not rendered on this host"
• "this surface is labs-only" · "the migrated host cannot do X"
• "the selector is gone" · "Flow removed it"
• a raise_if_migrated-style guard, a capability table, or a KNOWN_ISSUES entry asserting an absence
• an issue reply telling a user a feature cannot work

The rule this exists to enforce

A SELECTOR THAT DOES NOT MATCH IS EVIDENCE ABOUT THE SELECTOR.
IT IS NEVER EVIDENCE ABOUT THE FEATURE.

A wait that times out tells you your anchor missed. It tells you nothing about whether the thing exists. To claim absence you need a positive observation of absence — a DOM inventory that lists what IS there, a network log that shows what was and was not called. "It timed out" is not that.

Written from a two-line failure that cost a day. On 2026-09-06 gflow character create was believed impossible on flow.google.com. The entire chain came from one 20 s readiness timeout: selector missed → gate timed out → "no prompt textbox" → "labs-only surface" → "renders no prompt textbox for it, ever", shipped in a code comment, the CHANGELOG, a release ledger and a test class NAME. #701 then added a guard that aborted before probing the DOM — which made the claim unfalsifiable, because no run could ever look. Reality: the editor was fully present, on the same backend; labs renders React + Slate, the migrated host renders Angular + ProseMirror. Seven selectors changed and it worked. Thirty minutes of DOM reading would have prevented all of it.

Never put a guard in front of a probe. A fail-fast that runs before the evidence is collected deletes the evidence that would correct it. If you must fail fast, fail after looking, and say what you looked at.