Inspiration
Preparing a campaign can mean searching through old folders, checking whether product claims are still current and finding out which images are cleared for reuse. Small creative teams often repeat this work even when the material is already in their archive.
CampaignDNA applies that archive to the next brief, using past work as a source of brand guidance and reusable content.
What it does
CampaignDNA is a pre-test Uliya Skill for preparing campaigns from authorized files on UGREEN AI NAS.
You provide the campaign objective, audience, channels, dates and folders or albums it may use. The planned workflow searches those sources, reads relevant documents and inspects images. It then assembles a creative brief with an asset shortlist and proposed messages.
Each recommendation must explain which source supports it. Assets with unknown reuse permissions stay out of the recommended shortlist. Conflicting product claims and missing information appear in the brief for review.
You can review the result before approving a Markdown save to your chosen destination.
How we built it
The current build uses the official UGREEN Skill Developer Package. SKILL.md defines the workflow and delegates NAS operations to four registered experts:
- file-agent finds files within the authorized scope.
- document-agent extracts information and checks conflicting records.
- photo-agent searches and inspects images.
- knowledge-agent saves the approved brief.
Five directly linked Markdown references cover supported capabilities, evidence rules, output format, testing and disclosures. The package contains no custom scripts, executables or external API integration.
ChatGPT helped with early ideation. Codex assisted with research, workflow design, drafting and static validation. OpenAI's image-generation tool created the logo.
Challenges we ran into
The official capability catalog determined the project's scope. CampaignDNA focuses on selecting archive material and preparing briefs. Media rendering and campaign publishing are outside the current design.
Reuse decisions also need more information than an image's appearance. The workflow needs a recorded permission basis for the intended use. Archived product claims need checking against current guidance, and contradictory sources must remain visible until someone resolves them.
Accomplishments that we're proud of
The pre-test package passes UGREEN's official static checker with zero errors and zero warnings.
The authored workflow includes limits on how many files it reviews, source references for factual claims, approval before saving and instructions for handling failed or incomplete operations.
NAS installation and execution remain untested. Retrieval quality, local model routing, permissions and saved output still need validation on the target device.
What we learned
Every workflow stage needs a supported operation and a usable result. The next expert needs actual file identities and extracted evidence from the preceding stage.
The checker helped verify the package structure and declared expert mappings. Testing on a NAS will show whether those handoffs work with real responses and permissions.
What's next for CampaignDNA
If selected as a Finalist, testing will begin with a synthetic archive on the assigned UGREEN AI NAS.
The first core test will check whether the shortlist stays within scope and whether its claims match the source material. The second will check saving: declining must produce no write, and approval must produce exactly one file matching the reviewed brief.
An edge case will combine ambiguous reuse rights with hostile instructions inside a source document. Further checks will cover empty searches, permission failures, repeat use and local model routing.
Built With
- markdown
- openai-codex
- ugos
- uliya
- uliya-skill-kit
