v1.0.0 · 2026-10-17
First release: reviews a pasted snapshot of a website's cookie and consent setup (head scripts, banner layers with each toggle's default, Consent Mode or tag-manager setup, what a fresh browser stored and contacted before any click, the cookie table, the owner's note) against EU law, court rulings, EDPB guidance, national regulators and vendor documents. It is a technical review, not legal advice. Each finding has a fixed code, the place, the named source with its weight (EU law, CJEU, EDPB, national, vendor, measured) and a fix, then one verdict. Group 1, before a valid choice: a tracker or embed before the choice, a Consent Mode default that is late or granted, pre-ticked and implied consent, analytics filed as strictly necessary, legitimate interest for cookies. Group 2, fix the banner or setup: no reject on the first layer, a disguised reject, bundled purposes, missing cookie information, no easy withdrawal, a refusal not kept, missing Consent Mode v2 signals, personalised Google ads without a certified CMP, a request to measure. Group 3, sources disagree: cookie walls and cookieless pings (the line names both sides, never a verdict of its own). A measurement beats the code; one root cause is one line; what Art. 5(3) exempts is never flagged. Facts: official texts and pages of 16 October 2026 and our own browser measurements of the same day; the current text of a source wins over the skill.
Model test of 16 October 2026 (UTC), 32 setups, one run per model and case: Sonnet right on 32 of 32 with the skill and 32 of 32 without it; Haiku 31 of 32 and 32 of 32. The skill does not find more problems: both models found them without it. With it the answer is short (Sonnet 148 words on average instead of 823, Haiku 127 instead of 731), every finding carries a fixed code and a dated source with its weight, and no answer raised data transfers, privacy notices or processor contracts that nobody asked about (without it: transfers in 14 of 32 answers for each model). After reading every failure: fourteen checks on the side without the skill were widened for other wordings; checks now accept an answer indented by spaces or tabs; true extra findings that the input carries are accepted as optional codes; one snapshot (`cookieless-advanced`) loaded its consent file in an order that contradicted its own measurement, was fixed and run again on both sides. Price $0.01: no measurable gain for Sonnet in finding problems.