Phase 2: Content Curation

The digital-content-curator agent prepares raw inputs for downstream analysis. It transforms screenshots and transcripts into structured, analysis-ready formats while preserving all relevant content.

Sub-processes

1. Screenshot to HTML Mockup

The image-to-html skill reads each screenshot and produces a semantic, unstyled HTML file that captures all visible UI elements and text. Any PII visible in the original screenshot is replaced with fake equivalents.

Note: Converting screenshots to HTML also reduces the AI workload in later phases, as the model can parse structured text far more efficiently than images. This contributes to more sustainable AI use — see the DEFRA AI Playbook: Sustainability guidance for more detail.

Output: html/<name>.html

2. Transcript to Curated Transcript

The curate-transcript skill removes off-topic content (project management, scheduling, social chat) while preserving domain knowledge, application walkthroughs, and technical detail verbatim.

Output: transcripts/<name>_curated.txt

How to Run

Invoke the digital-content-curator agent via your AI coding assistant with the appropriate plugin or extension loaded. The agent discovers all raw files in the screenshots/ and transcripts/ directories and processes them automatically. See Tooling for setup instructions.

For example:

Please could you run the digital-content-curator agent to prepare all the screenshots and transcripts for analysis? Thank you.

What It Produces

• html/*.html — semantic HTML mockups of every screenshot
• transcripts/*_curated.txt — curated transcripts with off-topic content removed

Large File Sets

When there are many screenshots or transcripts (50+), the curator agent may exhaust its turn budget before finishing. See Content curation stalls on large file sets in the Troubleshooting section for a resumable bash workaround.

For full details on the available skills and tooling, see the Tooling section.