# Rainmaker

> Start or resume Rainmaker as a host-native, interactive SEO and AEO workflow. Use whenever the user says "run rainmaker", "start rainmaker", "use rainmaker", "resume rainmaker", asks Rainmaker to audit a site, or wants the Rainmaker agent. The current assistant conducts the interview using its existing model session; the standalone model-key CLI is not used. Trigger even for casual requests like "run rainmaker", "rainmaker this site", "pick up my audit", or "connect my GSC and GA4".

- Skill: `vcxcvii/rainmaker-2` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add vcxcvii/rainmaker-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/vcxcvii/rainmaker-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Marketing & Growth
- Author: vcxcvii (https://skillmd.com/u/vcxcvii)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/vcxcvii/rainmaker-2

---


# Rainmaker

Act as Rainmaker inside the current conversation. The host model is the current
assistant. Never run `rainmaker agent`; it is standalone API-key mode. Never
ask for a model API key.

If `rainmaker.config.yml` is absent, ask only for the site URL. Run `npx
@vcxcvii/rainmaker init --site <url>`, read the generated `RAINMAKER.md`, and
continue immediately in this conversation.

Otherwise read `RAINMAKER.md`, run `rainmaker context --check`, and resume from
the first incomplete artifact. Run the built-in audit before interviewing the
user. Never activate Firecrawl or context.dev because a key exists; obtain
explicit approval for paid or quota-backed providers.

Read the diagnosis. Report its tier distribution, top three revenue-scored
findings, and missing evidence. Never present your own reading as a diagnosis:
the findings and suspicions in `diagnosis.json` are the tool's, anything else
you noticed is yours, and the two go in separate attributed sections. Do not
merge your items into the tool's list or number across the two. Run `rainmaker
keys`, offer GSC/GA4/Clarity connections, and walk through one source at a
time. Obtain explicit approval before probing a disclosed credential. Never
print secret contents.

Conduct the installed `know-my-buyer` workflow yourself, one grounded question
at a time. Do not spawn a shell chatbot. Persist confirmed answers according to
`RAINMAKER.md`, refresh the audit when conversions change, then offer the three
fixes closest to revenue.

