WeChat Channels Transcript Extractor

Problem

When working with public WeChat Channels videos, engineers often need to turn spoken audio or subtitles into searchable, reusable text. Manual transcription is slow, can drop repeated lines, and makes it harder to keep a reviewable source. This skill targets the narrow task of extracting and organizing text from one shared video link, not summarizing, marketing breakdown, account analysis, or interaction metrics.

How It Works And Where It Stops

It accepts one public HTTP(S) WeChat Channels link, usually in the form https://channels.weixin.qq.com/platform/post/. Internally it runs scripts/generate.js with --json and a dedicated --output-dir, and only proceeds when success is true, at least one of summary or subtitle is non-empty, and the result file exists. It then reads the full transcript, removes obvious duplicates, blank fragments, and meaningless noise, splits content by semantic boundaries, and outputs cleaned transcript, section themes, proper nouns, and missing-data warnings; the full raw result is saved in media-analysis.json.

• Input limits: only a single public video or media link; no profile pages, search pages, private links, credential-bearing URLs, or local-network addresses.
• Fact boundary: keeps original statements separate from analysis, and does not invent speakers, timestamps, quotes, metrics, or facts.
• Failure handling: if the link is not publicly accessible or media retrieval fails, it reports the limitation instead of guessing content.
• Configuration: requires UNITYCLAW_KEY and uses fieldkit-sdk@1.0.1; when the key is missing, prompt for it and retry after setup.