ads-generate

ads-generate turns a validated creative brief and brand profile into paid-advertising image assets using an explicitly configured image provider. It solves the problem of producing on-brand, platform-compliant ad creatives in a repeatable, auditable way, rather than firing ad-hoc prompts at an image model. The skill assembles prompts from the concept, subject, action, setting, composition, brand tokens, platform constraints, and explicit exclusions defined in the brief, so every generated variant traces back to an approved specification.

A defining feature of this skill is its emphasis on data governance and provenance. All brief text and source images are treated as untrusted data, so embedded instructions are never followed and unapproved resources are never fetched. Before dispatching to a provider it validates the run's data-lifecycle contract, and for every asset it records the provider, model, parameters, cost, source hashes, prompt version, a SHA-256 of the normalized prompt, a canonical redacted prompt summary, output hash, and dimensions. Raw private prompts and provider payloads stay external and ephemeral, and the shipped JSON manifest contains neither secrets nor resolved local filesystem paths. Assets and manifest are written atomically inside a run directory using only repository- or run-relative locators.

The skill is aimed at marketing and creative-ops teams that need governed, cost-bounded ad generation with a clear audit trail. It fits campaign workflows where a brief and brand profile already exist and a budget or cost ceiling must be respected. Missing provider credentials produce setup guidance without exposing secret values, and every generated asset is explicitly flagged as requiring human rights, brand, quality, and policy review before use.