菜单
Interview the user's content-publishing goal, map it to a capability tier, recommend the minimum setup, then run the matching recipe from CATALOG.md. Use this skill for any "publish / grow / analyze on Douyin / XHS / X" request.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Optional polish + scoring of a draft before publish, and engagement scoring after.
Generate a Broadcast Kit content-batch contract from Markdown, a directory, a repo path, or a publish registry.
Join raw platform metrics JSONL back to registry/manifest metadata and compute experiment scoring priors.
Fetch and normalize platform metrics into the Broadcast Kit metrics snapshot shape.
Publish one Douyin manifest via the in-package Playwright publisher.
Publish a tweet or thread to X via NyxID proxy (preferred) or direct API (fallback).
| name | broadcast-pipeline |
| description | Interview the user's content-publishing goal, map it to a capability tier, recommend the minimum setup, then run the matching recipe from CATALOG.md. Use this skill for any "publish / grow / analyze on Douyin / XHS / X" request. |
Goal-first wrapper around broadcast-kit. Catch the user's goal in plain English first, then narrow the setup and recipe to fit. CATALOG.md is the reference menu, not the starting point.
Trigger this skill when the user says any of:
Open with one wide question. Do not paste the playbook template, capability tier table, or recipe list yet.
"What do you want to do with broadcast-kit? In your own words is fine."
Then ask at most 2-4 clarifying questions, scoped to what you still don't know:
Stop interviewing as soon as you can place the user on the tier table below.
| User signal | Tier | Recipe |
|---|---|---|
| "I have a finished video / image set, just publish it" | Tier 1 | E (with --video-file) or direct publish |
| "Polish this draft before I publish" | Tier 1 | A or C |
| "I have a publish-registry / inventory" | Tier 2 | B-style with registry-to-manifest |
| "I have a PDF / markdown, make the video" | Tier 3 | E (full source-to-video) |
| "Grow my account over N weeks" | Sprint | B with playbook |
| "Why did my last posts underperform" | (any) | analyze_misses + engagement_score |
Tier definitions live in CATALOG.md → Capability tiers.
Pick the smallest setup that satisfies the goal. Do not ask the user for NotebookLM or SlideSync credentials unless their goal lands on Tier 3. Likewise, skip BITGRIT_API_KEY / higgsfield prompts unless they asked for virality scoring.
broadcast-kit setup --for tier1 # publishers + LLM only
broadcast-kit setup --for tier2 # tier1 + publish-registry path
broadcast-kit setup --for tier3 # tier1 + NotebookLM + SlideSync
For Sprint, run setup --for tier1 (or tier2 if they have a registry) plus write state/playbook/<platform>.yaml from the interview answers. For Analysis-only, setup --for tier1 is enough — they already have data, they just need LLM creds.
Always end the setup step with:
broadcast-kit doctor
and confirm the relevant ok_for_* flag is true before running the recipe.
If doctor fails, fix the gap (login, missing binary, missing env var) before moving on. Don't proceed and hope.
Pull the recipe straight from CATALOG.md — don't re-copy it here. Anchors:
--video-file)broadcast-kit registry-to-manifestanalyze_missesRead CATALOG.md only at this point, and only the section you need.
content_brain.publish_decision == "hold" or reviewer.recommend_publish == False. Surface the reasoning to the user and ask.virality_score returns status="skipped", which is correct.state/. Auth cookies live there; it is gitignored.CATALOG.md — Capability tiers — full menu of parts and recipesdocs/internal-onboarding.md — step-by-step for internal staffdocs/optimizers.md — optimizer deep-divedocs/publishers/ — per-platform manifest shape and success contracts