Comment Analysis Studio

The problem it solves

When writing short-video or live-commerce copy, teams often lack reliable audience language. Objections, purchase concerns, common Q&A, and spoken hooks can drift into generic platform phrasing if they are invented instead of grounded in real comments. This skill turns a single post, a named category, or pasted comments into reviewable audience-language material with source labels, ready for scripting, live replies, and FAQ drafting.

How it works

The core output is a grouped brief covering objections, verbatim quotes, common Q&A, and spoken hooks. By default, it reads one page of top-level comments and does not expand nested replies. If the user has already pasted comments, the brief can be produced without paid lookups. If the user explicitly requests reading comments from Douyin, TikTok, Xiaohongshu, Instagram, YouTube, or X, the skill first explains the target, operation, credit price, and number of queries before running them.

Key steps include:
- Scope the source: collect comments from a named post, a named category page, or user-pasted content.
- Label provenance: mark looked-up items as found with the read date; mark pasted items as user-provided.
- Extract material: trace each objection, FAQ, and hook back to source comments where possible.
- Deliver the brief: include a short read, quote list, common Q&A, live answers, and hooks; if paid lookups ran, also include task_id, terminal state, and billing.net_charged_credits.

Remote calls use the bundled client. Each paid request uses a stable client_request_id to avoid duplicate submissions. On query failure, inspect error.code and error.message first, and change parameters only when the message points to a field or ID.

Boundaries

This skill fits the pattern of one post plus one comment page plus one grouped brief. It is not meant to blind-scan the whole site, start from a large unscoped post set, or replace full account diagnostics, viral-video remakes, or complete live-stream scripts. If a platform lacks comment-query support, it falls back to pasted content. Missing counts should be reported as missing rather than estimated from similar posts, and inferences should be labeled as such with supporting evidence.