RankRaft skill for AI coding agents - RankRaft AI Knowledge Base

The RankRaft skill is a small instruction pack for AI coding agents. With it installed, an agent asked to “pull our visibility into BigQuery” or “add these prompts from a CSV” writes working code the first time: it uses the right SDK, the right endpoints and filters, handles pagination and rate limits, and knows when you should use the MCP server instead of code. It follows the open Agent Skills format: a SKILL.md file plus a few reference pages, all plain Markdown. Any agent that reads skills can use it.

Want to ask questions about your data in a chat app rather than write code? You don’t need the skill — connect the MCP server instead.

Install

Get the skill once:

git clone https://github.com/RankRaft/rankraft-skill

Then put it where your agent looks for skills:

Recommended: the plugin, which installs the skill and keeps it updated.

/plugin marketplace add RankRaft/rankraft-claude-plugin
/plugin install rankraft@rankraft

Or from a shell: claude plugin marketplace add RankRaft/rankraft-claude-plugin, then claude plugin install rankraft@rankraft. Scopes, updates and removal: Claude Code plugin.Or copy the skill:

# this project only
cp -r rankraft-skill .claude/skills/rankraft
# every project
cp -r rankraft-skill ~/.claude/skills/rankraft

In Claude on the web or desktop, custom skills are uploaded as a zip:

Zip the folder: cd rankraft-skill && zip -r ../rankraft-skill.zip . -x '.git/*'
Open Settings → Capabilities → Skills and upload rankraft-skill.zip.
Make sure code execution is on; skills run inside it.

On Team and Enterprise plans, an owner may need to allow custom skills first.

Codex (CLI and IDE extension) loads skills from your skills folder:

# every project
mkdir -p ~/.codex/skills && cp -r rankraft-skill ~/.codex/skills/rankraft
# this project only
mkdir -p .codex/skills && cp -r rankraft-skill .codex/skills/rankraft

Restart Codex. Ask “what skills do you have?” to check it is listed.

For code you’ll run yourself, add the skill to a ChatGPT Project:

Create a project (for example “RankRaft integration”).
Paste the contents of SKILL.md into the project’s Instructions.
Upload the files in references/ as project files.

Every chat in that project then follows it. For coding inside your repository, use the Codex tab instead.

# every project
mkdir -p ~/.gemini/skills && cp -r rankraft-skill ~/.gemini/skills/rankraft
# this project only
mkdir -p .gemini/skills && cp -r rankraft-skill .gemini/skills/rankraft

Restart Gemini CLI so it picks the skill up.

mkdir -p .cursor/skills && cp -r rankraft-skill .cursor/skills/rankraft

Cursor also reads skills from .claude/skills/, so a copy installed for Claude Code works here too.

Copy the folder wherever your agent reads SKILL.md files. If it has no skills support (Windsurf, Cline, GitHub Copilot and others), paste SKILL.md into the project’s rules or custom instructions and keep references/ in the repository so the agent can open it.

The skill repository

github.com/RankRaft/rankraft-skill is the skill itself: plain Markdown and one shell script, MIT licensed, nothing specific to one agent.

rankraft-skill/
├── SKILL.md                    # the entry point every agent reads
├── references/
│   ├── api-reference.md        # request and response shapes, filter fields, limits
│   ├── errors.md               # every error code and what to do about it
│   ├── recipes.md              # ready-made tasks and curl equivalents
│   └── why-the-sdks.md         # when calling the API by hand is worth it
└── scripts/
    └── smoke.sh                # checks a key, its access and your site ids

File	The agent reads it when
`SKILL.md`	starting any RankRaft task: REST or MCP, SDK quick starts, the rules that cause most errors
`references/api-reference.md`	it needs exact request or response fields, or the filter fields an endpoint accepts
`references/errors.md`	handling or explaining an error
`references/recipes.md`	building something common: visibility over time, leaderboards, citation share, CSV prompt loads, warehouse exports, long backfills
`references/why-the-sdks.md`	deciding whether to use an SDK or raw HTTP
`scripts/smoke.sh`	checking that a key works

The Claude Code plugin packages this same directory, synced automatically, so both stay identical.

What it teaches

REST or MCP: code that pulls data or manages prompts uses the REST API and an SDK (Enterprise); questions in a chat app use the MCP server (every paid plan).
The SDKs: rankraft-js and rankraft-py, with working quick starts.
Every endpoint, with the filter fields each report accepts and the limits that apply.
The rules that cause most errors: server-side only; the key goes in a header; dates are inclusive UTC days; cursors belong to one query; filter values are strings; unknown fields are rejected; prompts are disabled, never deleted.
Every error code and what to do about it.
Recipes: visibility over time, a competitor leaderboard, citation share by domain, loading prompts from a CSV, daily crawler traffic into a warehouse, and long backfills.

It also includes scripts/smoke.sh, which checks a key, its access level and your site ids in one command:

RANKRAFT_API_KEY=rr_api_… bash scripts/smoke.sh

Try it

With the skill installed, ask your agent something like:

“Write a script that exports our daily Visibility Score by engine for the last 90 days to a CSV.”
“Add the prompts in prompts.csv to our site, as a dry run first.”
“Load yesterday’s AI crawler traffic into Postgres every morning.”

Staying up to date

The skill follows the API’s OpenAPI specification.

Claude Code plugin: turn on auto-update once, or run claude plugin marketplace update rankraft and claude plugin update rankraft@rankraft. See Keep it up to date.
Copied install: run git pull in your rankraft-skill clone and copy the folder again.

Source

github.com/RankRaft/rankraft-skill · Claude Code plugin: github.com/RankRaft/rankraft-claude-plugin · MIT licence

​Install

​The skill repository

​What it teaches

​Try it

​Staying up to date

​Source

Install

The skill repository

What it teaches

Try it

Staying up to date

Source