The ads skill output was missing structural enforcement for Google RSA
platform limits, causing AI agents to generate non-compliant ads:
- variable headline/description counts (not exactly 15/4)
- character lengths exceeded (>30 chars headlines, >90 chars descriptions)
- missing negative keyword lists
- missing ad group structure labels
This commit adds a 'Google RSA Output Spec' section that:
- Enforces 15 headlines (≤30 chars) / 4 descriptions (≤90 chars) per RSA
- Requires explicit ad group structure, sitelinks (≥4), callouts (≥4)
- Requires labeled negative keyword list (≥8 entries)
- Specifies mandatory output ORDER (Negatives + sitelinks before RSAs)
so they survive output truncation in long generations
- Adds CFM (Brazilian medical regulation) forbidden-term list for
pt-BR medical practice contexts
- Adds a pre-response self-check checklist
Validated on a marketing-eval harness across 3 medical sites: scores
moved from 47/47/20 (all critical-fail) to 100/100/87 (all critical-pass)
after applying the spec.
Co-authored-by: Erion De Andrade <231011322+erionjuniordeandrade-a11y@users.noreply.github.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
erionjuniordeandrade-a11y
13fa45674c