Content Learning Remix

Problem

When research notes, articles, social posts, audio, and video are mixed together, the issue is often not “summarize this,” but broken evidence. Source claims, timestamps, subtitles, visual cues, and model inferences can become hard to separate. This skill first converges heterogeneous input into a canonical content.json evidence package, then uses that package to produce learning, research, or original creation outputs.

How It Works

It follows a five-layer pipeline: Route detects the source type; Acquire extracts metadata, body text, subtitles, or transcripts, and samples frames when visual evidence matters; Understand normalizes the material while preserving timestamps and provenance; Create builds a script, article, lesson, or platform adaptation from mechanisms and ideas; Deliver selects a complete bundle for the target audience. For audio and video, it prefers creator-provided subtitles or automatic captions before falling back to ASR. For visually important content, it can sample frames with --frames 8 and reason over demonstrations, slides, products, or editing cues.

Boundaries

It is useful for public articles, media links, local files, and pasted text, especially when the goal is remixing source material into a lesson, script, article, or platform-specific piece. It does not guarantee access past login walls, region blocks, anti-bot controls, or expired signatures. ASR output should be treated as evidence rather than publication-ready copy, and visual claims should not be made unless frames were actually inspected.