CRO Actions

Cutting a form from 11 fields to 4 lifted conversions 120%. Cutting one from 9 to 6 dropped them 14%. Both numbers are real, both are published, and for a while I was quoting whichever one suited the slide. That is what most CRO advice is: someone else's result, abstracted until every detail that made it true has been stripped out, handed back to you as a rule. "Reduce form fields" is not a rule. It is a thing that happened twice, in opposite directions. cro-actions is the middle link of my CRO chain. cro-audit finds what is broken, this skill decides what to do about it, and cro-optimizer reads the result after it ships. Its job is to turn a ranked finding into a plan a team can actually execute — and it refuses to emit an action that is not concrete. Every action must carry two things or it does not ship: real evidence (a source-cited measured lift, or an explicit unverified flag) and a real target (a named SpiderPublish sys-* component to install, or a specific content edit with a named handoff). No evidence and no target is not an action. It is a vibe. Three scripts enforce that on every plan. One rejects any action without a cited metric and caps a plan at 20% unverified. One grades the effort estimate against its change-type band, so "rebuild checkout" cannot be filed under hours. One lints for empty verbs — "improve the CTA" fails, because it names no element, no change and no success metric. It also knows what not to promise. Individual lifts are never summed into a bundle projection, because effects overlap. It will not stack more than 2-4 component installs on one page, because uplift plateaus and components fight for the viewport. On regulated pages it refuses manufactured urgency and fake social proof outright. Guidance-only. 32 action blueprints, 6 references, 7 fix recipes, 10 learnings, 4 scripts, zero methods. SpiderPublish installs and deploys, copy-engine writes, section-designer rebuilds, media-generator renders. This skill decides and sequences.