6.0.0: AI-generated ads only, research + questions before generation, 2026 feed layer, creative diagnostics

[aievolutionpl]

What this adds

This release teaches the skill what current Meta ads look like, which structures persuade, and what changed on the platform in 2026. Sources: Behance and Dribbble portfolio patterns (from a Behance "meta advertising" screenshot), Penji's Meta design guidance, Meta's Ads Guide, and about 30 sources from 2026 on creative performance, trends and placement changes. Figures from third parties are labelled as such and never presented as promised results.

Behance, Dribbble, Penji and facebook.com were blocked by the research environment's network proxy. Their content came from search summaries and the supplied screenshot, and the docs say so.

New rules (additive, no renumbering)

• R45 · Commit to one visual language. A creative uses one style: its type, colour, image treatment and one signature device. Brand identity outranks trend. Imperfection has to come from a real process, not a filter. A portfolio board is not an ad.
• R46 · Format follows proof and funnel. Choose the persuasion skeleton before the style. A format is only available if its proof exists (no verified number → no stat drop).
• R47 · Diversity is delivery. Andromeda groups near-duplicate ads. A campaign set should vary persona, motivation, hook, format and style. Controlled tests still rotate one variable.
• R48 · Design for the 2026 placement system. Covers the unified Reels/Stories safe zone (top 14%, bottom 35%, sides 6%), 4:5 export at 1440×1800, text truncation, Advantage+ enhancements (expansion, generated backgrounds, Muse Image) and "AI info" labels. Adds a new hard fail, R48-fake-ui.

New references and examples

• references/style-atlas-2026.md: 12 visual languages. Each has its DNA, fit, a prompt fragment and its typical failure. Also covers format pairings, "trend slop", and how to read reference boards (with a worked teardown of the Behance page).
• references/static-ad-formats.md: 15 static formats. Each has a 4:5 skeleton, copy budget, proof requirement and integrity rule. Also covers funnel fit, format by vertical, hook-to-format pairings and carousel sequencing.
• examples/07-2026-style-directions.md: 5 fictional briefs pairing a format with a style, each with a prompt and typesetting handoff, plus a concept campaign set built for Andromeda.
• assets/style-atlas-2026.png plus source assets/generated/style-atlas.html: six fictional ads typeset in HTML/CSS. No image model was used; provenance is documented in examples/06-readme-showcase.md.

Behaviour change in scripts/qa.py

• The 9:16 keep-out now follows Meta's unified zone. It is expressed as a share of canvas height and measured from the edge, and the stricter of margin and chrome wins. The old code added fixed 250/320px values on top of the 86px margin. That over-restricted the top and let a CTA sit in the Reels action area at the bottom (e.g. y 1300–1420 used to pass).
• The format check now also accepts 1440×1800 (4:5) and 1440×2560 (9:16).
• Three new self-tests cover this.

Updated

Platform compliance (§6), variation matrix (§6), performance loop (fatigue signals), hook engineering, anti-slop (§1b 2026 tells), niche playbooks, QA vision pass (visual_language, fake_functional_ui), layout system, core.md §20, design-rules.md, SKILL.md routing, and both READMEs (PL + EN) with the new board.

Validation

• python scripts/check_docs.py: PASS, 37 markdown files
• python scripts/test_qa.py: 16/16 cases passed

🤖 Generated with Claude Code

https://claude.ai/code/session_01VhzbuY17VENNVTgxod4KW3

---

Generated by Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.