xhs-workflow

Personal Xiaohongshu (小红书 / RED / Little Red Book) workflow skill for consistent content production across agents and computers. Use whenever the user asks for Xiaohongshu strategy, raw-video-to-post work, idea development, draft editing, photo/screenshot carousel planning, titles, cover text, captions, scripts, tags, first comments, publishing bundles, or post diagnostics. First route the request by input type, then follow the matching workflow. Mandatory gates: final copy must pass the xhs-copy humanizer gate (`dependencies/xhs-copy/humanizer/`), and final visual deliverables must pass the visual quality workflow. Preserves the existing XHS strategy rules for platform-tuned titles, body templates, tags, sensitive wording, pre-publish checks, diagnostics, and staged video production.

Rewrite Xiaohongshu copy so it sounds like a real person, not AI, a public account, or a generic marketing script. Use as the mandatory copy gate for titles, cover text, captions, scripts, carousel text, first comments, and interaction prompts.

The XHS copy sub-agent for the xhs-workflow Director. Owns all text production for Xiaohongshu — titles, body, photo-carousel copy, opinion/narrative copy, tags, engagement hooks, and the final de-AI humanizer gate. The Director (xhs-workflow/SKILL.md) routes here whenever copy must be written or edited; it passes the brief + confirmed direction, and this skill returns ready-to-publish wording. Platform strategy (CES, cold-start, recommendation rules) and the compliance red-line map stay in the Director — this skill applies them, it does not redefine them.

The XHS image/cover sub-agent for the xhs-workflow Director. Produces Xiaohongshu (小红书) cover images and in-post images in two modes — EDIT mode (turn a user's REAL photo into a hit-style cover while preserving their real face/identity) and GENERATE mode (create a cover/image from scratch with text-to-image when there is no real person to preserve). The wording on the image (title/copy) comes from xhs-copy via the Director; this skill owns the image itself. Always route by "is there a real face that must stay the real person?" before choosing a mode.