marketing-skillsSAFE

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: ad-creative-generation
description: Generate on-brand ad creatives — visuals + copy — for Google, Meta (Facebook / Instagram), and other paid platforms via the Hyper MCP. Extracts brand identity from a website, writes ad copy variants, and produces brand-consistent images using reference-based image generation. Use when the user asks for ad creative, ad copy variants, RSA headlines, Meta ad creative, display ads, carousel ads, or A/B test variants.
metadata:
  version: 1.0.0
requires_toolkits:
  - image_gen
icon: image_gen
short_description: Brand extraction, ad copy variants, and on-brand images for Google and Meta placements.
---

# Ad Creative Generation

Generate ad creatives — both visuals and copy — that match a brand's identity. This skill orchestrates brand extraction, copywriting, and image generation into a single workflow.

## Requirements

This skill assumes the [Hyper MCP](https://app.hyperfx.ai/mcp) is connected to your agent so the tools below are available. Brand extraction also requires Firecrawl to be configured under your Hyper integrations.

### How to run the tools in this skill

Every tool in this skill is named by its canonical tool name. Run it with the call your surface gives you:

| Surface | Find a tool | Run it |
| --- | --- | --- |
| MCP client (Claude, Cursor, Codex, ChatGPT) | `search("<what you want to do>")`, then `describe("<name>")` | `call("<name>", {...})` |
| Hyper CLI | `hyperai search "<what you want to do>"`, then `hyperai describe <name>` | `hyperai call <name> --json '{...}'` |

If a tool is not found, its integration is not connected or not enabled for the workspace: stop and tell the user which integration to connect.

## Tool surface

| Group | Tools |
|-------|-------|
| Brand extraction | `firecrawl_branding_extract` |
| Archetype recommendations (conditional) | `cmo_ad_brief` |
| Competitor ad tracking (conditional) | `cmo_ads_track`, `cmo_ads_tracked_list` |
| Image generation (default) | `images_generate` (`model="gpt-image-2"`) |
| Image generation (text-heavy) | `images_generate` (`model="nano-banana-pro"`) |
| Image generation (photoreal product shots) | `images_generate` (`model="seedream-4.5"`) |

The `cmo_*` tools are conditional — they appear when the CMO toolkit is
enabled in the workspace. When absent, skip them and use the reference
frameworks; do not fail the workflow.

For deeper image-tool selection guidance see the `image-generation` skill. To turn finished creatives into running campaigns see `google-ads` and `meta-ads`.

## Out of scope

- Picking the right image model when the task isn't ad creative — use `image-generation`.
- Creating, launching, or budgeting actual ad campaigns — use `google-ads`, `meta-ads`, `tiktok-ads`, `pinterest-ads`, or `amazon-ads`.
- Searching live competitor ads — use `meta-ads-library`; to save and monitor competitor ads cross-platform (Meta + Google Ads Transparency, durable creatives, longevity ranking), use `cmo_ads_track` / `cmo_ads_tracked_list` when the CMO toolkit is enabled.

## Critical rules

- **Always extract branding first** before generating visuals for a website/brand.
- **Always pass both logo and screenshot** as reference images — SVG logos are auto-converted to PNG.
- **DO NOT display image URLs** — they are automatically shown in chat.
- **Read `brand.screenshot.description`** to understand the product — never guess from the company name.
- **Default to logo + headline + product screenshot** for SaaS / product ads unless the user requests a different style.
- **Use `model="gpt-image-2"` for the initial ad creative by default** — with reference images for branded work, or without for loose first-pass concepts.
- **Do not default to `model="nano-banana-pro"` for the first creative pass** — use it when the user explicitly asks or when readable text inside the image is the main requirement.
- **Set `aspect_ratio` to the placement** — `1:1` (feed), `9:16` (story), `16:9` (banner) — and `quality` (`"draft" | "standard" | "high"`); `images_generate` takes those, not raw pixel sizes.
- **Respect character limits** — Google RSA headlines are 30 chars, descriptions 90 chars; Meta primary text is 125 chars visible.
- **Change one variable per variant** so test results are attributable.
- **Match aspect ratio to placement** — feed is 1:1, story is 9:16 (use `1024x1536`), banner is 16:9 (use `1536x1024`).

## Routing Table

Based on what the user needs, read the appropriate reference file:

| User Need | Reference |
|-----------|-----------|
| "Extract branding from this site" / brand colors, logo | `references/brand-extraction.md` |
| "Write ad copy" / headlines / variants / hooks | `references/ad-copy-frameworks.md` |
| Google RSAs / display ads / Performance Max assets | `references/google-ads-creatives.md` |
| Meta / Facebook / Instagram ads / carousel / stories | `references/meta-ads-creatives.md` |

When the task spans multiple areas (e.g., "create ad creatives for this website"), follow the full workflow below.

## Core Workflow

### Phase 1: Brand Extraction

Call `firecrawl_branding_extract` with the website URL. This single call returns branding data, a saved logo, and a website screenshot:

```python
brand = firecrawl_branding_extract(url="https://example.com")
```

The result contains:
- `brand.logo.file_id` — logo image to use as reference
- `brand.screenshot.file_id` — website screenshot to use as reference
- `brand.branding` — colors, typography, spacing, personality

Both `logo` and `screenshot` are saved images ready to pass as `reference_images` to image generation tools. SVG logos are automatically converted to PNG, so always include the logo — no need to filter by format.

**For deeper guidance on brand extraction, read `references/brand-extraction.md`.**

### Phase 2: Ad Copy Generation

**If `cmo_ad_brief` is available, call it first** — it returns ranked ad
archetypes for the brand's goal and platform with the reasoning, hook
examples, structure beats, d

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (5)

.prettierignore

Once a day the user says: "Run the cold-email cadence — send any due touches and pull replies."

2. **User explicitly asks to be ignored**: If the user says something like "ignore this", "don't respond", or similar.

2. When the user mentions a niche or competitor channel, run `research_top_thumbnails` first — never guess what's working.

> **Warning**: This tool requires a **Page Access Token** and will fail with error 190 ("This method must be called with a Page Access Token") when using a User Access Token. If this happens:

Gates applied: no_behavioural_pass.