菜单
Assemble a funnel roast report card from an existing run directory (funnel.json, walk artifacts, and analysis files). Use when writing the roast report, report card, or Funnel Score HTML after a walk. Do not use for capturing a site, fetching ads, starting Niko, or running npm run score.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/Texarkanine/rmf --skill write-the-roast命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Run a full funnel roast: collect a sales-funnel URL and a Meta or Facebook ads deeplink, fetch the ad, walk the live funnel with Playwright, apply the matching marketing rubric at each hop, and write the report card. Use when starting a roast, roasting a funnel, or producing a roast report from an ad plus a landing page. Do not use for an isolated Ad Library fetch, an isolated walk, Niko, or npm run score.
Run the Playwright capture CLI against a live sales-funnel URL, write steps.json, and print that hop list. Use when walking a funnel, capturing checkout, or checking whether a store reaches payment fields. Do not use for scoring, roasting ads, writing report cards, or starting Niko.
Extract Meta or Facebook Ad Library ads from a deeplink into ads.json and downloaded creatives. Use when fetching an Ad Library id, a library search or page, ad copy, creative URLs, or ad media onto disk. Do not use for roasting a funnel, scoring ads, writing funnel.json, or later roast steps.
基于 SOC 职业分类
正在显示 SKILL.md
| name | write-the-roast |
| description | Assemble a funnel roast report card from an existing run directory (funnel.json, walk artifacts, and analysis files). Use when writing the roast report, report card, or Funnel Score HTML after a walk. Do not use for capturing a site, fetching ads, starting Niko, or running npm run score. |
Single-agent workflow. Disk is the handoff. Chat output is the report path.
Do not browse. Do not recapture. Do not drive a browser. Do not call npm run score. Do not invent pages, ads, or findings.
flowchart LR
workdir["Collect workdir"] --> layout["Read report-layout"]
layout --> facts["Read on-disk facts"]
facts --> html["Write report.html"]
html --> print["Print report path"]
Resolve one value: the roast run directory (the path /start-the-roast created, or --workdir <path>).
If it is missing or not a directory, ask for it and stop. Do not write HTML.
Read references/report-layout.md. Do not write the report until this read is done.
Read only files that exist under the workdir:
funnel.jsonads.jsonwalk/steps.jsonanalysis/walk/artifacts/…, media/…)A missing file is a gap. Record it. Do not invent its contents.
Write <workdir>/report.html to the layout from Step 2.
Every roast line and every grade cites an artifact path that exists. Shared destination URLs get one journey tile. Do not include an Ad Library chrome screenshot.
If analysis/lp-mock.md names media/lp-mock.png and that file exists, put the image next to the primary ad / destination gap. Omit the image when the file is missing.
Print the absolute path of <workdir>/report.html. That path is the deliverable.
Do not paste report.html into chat. Do not paste funnel.json or steps.json.