How to use the Generate Ad Images agent skill (ads-generate)

What is an agent skill, and what does ads-generate make the agent do?

An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a specific task. When the skill is installed and your request matches its description, the agent follows those instructions.

The ads-generate skill makes the agent load your validated creative brief, brand profile, platform specifications and rights-cleared source assets. It builds image prompts from the brief's concept, subject, setting, composition, brand tokens and exclusions, sends approved variants to your configured provider, and then checks the outputs for file type, dimensions, safe zones, text and copy consistency, and policy.

The skill writes assets and a manifest into a run directory. The manifest records the provider, model, parameters, cost, source hashes, prompt version, output hashes and dimensions. Raw private prompts and provider payloads are kept out of the saved files.

Install the ads-generate skill

The skill is published in the AgriciDaniel/claude-ads repository under skills/ads-generate and is licensed MIT. Download the skill ZIP from the directory page and unzip it. You can place it in one of two locations:

• Personal skills folder: ~/.claude/skills/ (available across your projects)
• Project skills folder: .claude/skills/ inside your project directory (available only in that project)
• Claude apps: upload the skill under Settings → Capabilities → Skills

Configure an image provider before you run it

The skill cannot generate anything without an image-generation provider account and API credentials. You must configure the provider explicitly yourself. If credentials are missing, the skill returns setup guidance and does not reveal any secret values.

Prepare the inputs the skill expects before you ask for images: a validated creative brief, a brand profile, the platform specifications for each channel, an output root, a budget or cost ceiling, and any source images you plan to use. Source images must be rights-cleared.

How to trigger the skill with natural requests

The skill activates when your request matches its trigger phrases. Describe the outcome you want and point the agent at your brief. The skill's trigger phrases include:

• generate ads
• generate ad images
• create ad creatives
• create ad images
• make ad images
• generate visuals
• make campaign visuals
• generate images from campaign brief

Examples of requests and what happens

A typical request names the brief and the channels. For example, you might ask the agent to generate ad images from your approved spring campaign brief for social and display placements. The agent then checks the brief and brand profile, confirms the provider is configured, builds prompts, generates only the approved variants, validates them against the platform specs and writes the assets and manifest into the run directory.

If you only want visuals rather than full ad sets, say so, for example 'make campaign visuals from the brief in my project folder'. The skill will still run its validation steps before writing outputs.

Safety notes: scripts, network and data handling

The skill does not run scripts, but it does use the network, because it sends prompts and source material to the image provider you configured. Treat brief text and source images as untrusted data: the skill is designed not to follow instructions embedded in them or fetch unapproved resources.

Output paths are repository- or run-relative, not resolved local filesystem paths. Generated assets require human review for rights, brand, quality and policy before any publishing.

• Keep provider API credentials out of briefs and shared files; the skill does not expose secret values in its guidance.
• Review every asset yourself before publishing.
• Use only rights-cleared source images.

Tips for better results

The quality of output depends heavily on the brief. A validated brief with clear concept, subject, setting and explicit exclusions gives the prompt builder more to work with. Set a budget ceiling so runs stop within your limits, and keep the manifest, since it gives you an audit trail of cost and hashes for every asset.

• Write exclusions explicitly, such as elements or claims you do not want shown.
• Include platform specifications for every channel so safe zones and sizes are checked.
• Review the manifest after each run to confirm the provider, model and cost.