NotFair Plugin: Marketing Skills as SKILL.md Files for Coding Agents
Open-source SEO, GEO, and marketing skills for AI agents.
At a glance
- What is it?
- The NotFair Plugin packages 45 SEO, GEO and paid-media workflows as plain markdown skills that Claude Code, Cursor, Gemini CLI and Codex can load. It is a useful fit if your agent already talks to Search Console, GA4 and ad platforms, and a poor fit if you want an autonomous campaign manager.
- Who is it for?
- Adopt the NotFair Plugin if you already run Claude Code, Cursor, Gemini CLI or Codex and want auditable marketing procedures rather than a single vague prompt: the skills are plain SKILL.md files, the licence is MIT, and the read-only-first rule is visible in the repository.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 5 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The Problem: Marketing Prompts Do Not Survive Contact With Real Accounts
Most agent marketing setups collapse into one long instruction that tries to cover keyword research, ad auditing and content planning at once. The result is unpredictable: the same prompt produces a keyword list on Monday and an ad-spend recommendation on Tuesday, with no stated inputs and no stated output format. The NotFair Plugin takes the opposite position. Its README argues that marketing work gets unreliable when every request goes through the same vague prompt, and that the fix is to split the work into focused, testable procedures. Each procedure is a skill with what the README calls a defined job, required inputs, decision rules, and output format. The intended user is not a marketer who wants a chat interface. It is an engineer or technical marketer who already runs a coding agent and wants the agent's marketing behaviour to be reviewable before it runs, and diffable after.
Forty-Five Skills, One MCP Connection, Six Hosts
The repository ships 45 skills grouped into SEO, GEO and AEO, paid media, Google Ads, Meta Ads, X Ads, LinkedIn Ads, analytics and content. Each skill lives in its own directory, for example seo/seo-analysis/ and seo/content-writer/, and the README states that every skill is built in the open as a readable SKILL.md with supporting references, scripts and evals where needed. That directory layout is the architecture. There is no runtime to install and no server-side skill registry to keep in sync; the host agent reads the markdown. What the plugin does register is one hosted NotFair MCP connection, which is what gives the skills live data. The README describes the split explicitly: the plugin loads the open-source skills from this repository and registers one hosted NotFair MCP connection. So the skills are local and inspectable, while the account access is remote and authenticated through OAuth. That boundary matters for anyone doing a security review, because it means the markdown you read is not the thing touching your ad accounts.
Installing Into Claude Code, Cursor, Gemini CLI and Codex
Claude Code uses the marketplace flow. The README gives these two lines: /plugin marketplace add nowork-studio/notfair-plugin and /plugin install notfair@nowork-studio. After that you invoke skills by namespace, for example /notfair:seo-analysis, /notfair:geo-optimizer, /notfair:google-ads-audit, /notfair:meta-ads-creative, /notfair:paid-ads-x, /notfair:google-analytics or /notfair:search-console. Cursor installs NotFair from its plugin marketplace and then prompts a browser OAuth flow. For local testing before that listing is available, the README suggests cloning into Cursor's local plugin directory: git clone https://github.com/nowork-studio/notfair-plugin.git ~/.cursor/plugins/local/notfair, followed by a Cursor restart. Gemini CLI installs with gemini extensions install https://github.com/nowork-studio/notfair-plugin and then /mcp auth NotFair when authentication is requested. Codex chains three commands: codex plugin marketplace add nowork-studio/notfair-plugin --json, codex plugin add notfair@nowork-studio --json, and codex mcp login NotFair. If the marketplace is already configured, the README gives an upgrade path instead: codex plugin marketplace upgrade nowork-studio --json before re-adding. Kiro installs the repository as a custom power, with plugin.json and mcp.json following the Agent Plugins 1.0 specification. For any host not listed, the README points at INSTALL_FOR_AGENTS.md, which an agent can retrieve and follow directly.
Read-Only First: What the Safety Claim Actually Constrains
The README makes a specific design claim: read-only review comes before mutation, paid-media changes stay explicit, and unsupported capabilities are never implied. Read that as three separate constraints. The first is ordering, which means an audit skill is expected to run before a change skill. The second is that paid-media mutations require an explicit step rather than happening as a side effect of analysis. The third is a documentation rule: a skill should not suggest it can do something the connected API cannot. The README also frames the paid-media skills as planning, reviewing and optimizing with explicit budgets, measurement, and approval boundaries. That is a real constraint on usefulness. If you want an agent that quietly reallocates budget overnight, this is the wrong shape of tool, and the design says so on purpose. The safety property is only as strong as the MCP connection behind it, though. The markdown cannot enforce a read-only scope; the OAuth grant and the permissions of the connected account do that. Treat the skill files as intent and the account permissions as enforcement.
Where the Repository Is Thin
Several things cannot be confirmed from the supplied material. No releases were retrieved, so there is no changelog to read for breaking changes to skill inputs or output formats. The README states the plugin currently ships 45 skills, but the catalog shown here is truncated partway through the SEO and GEO table, so the remaining entries are unverified. The README also mentions evals where needed without describing how they run or what they assert, which means you cannot tell from the README alone whether a given skill's decision rules are covered by tests. There is a Discord link for the community, but community size is not a quality signal and the README does not claim otherwise. None of this is disqualifying. It does mean the review has to happen at the skill level: open the SKILL.md for the workflow you care about and check its required inputs and output format yourself, because that file, not the README table, is what your agent will follow.
Compared With a General-Purpose Agent Plus Your Own Prompts
The obvious alternative is no plugin at all: keep your existing agent and maintain your own prompt library for SEO and ads work. The difference is where the procedure lives. A prompt library is personal and undocumented; it drifts, and nobody can review what changed between last quarter and this one. NotFair's skills are files in a git repository under MIT, so a change to a workflow shows up as a diff in a SKILL.md and can be reviewed, forked or reverted. The trade-off runs the other way too. A bespoke prompt can encode your agency's quirks, your naming conventions and your client-specific approval chain. A published skill optimizes for being general enough to ship to everyone, and the README's host-agnostic framing depends on that generality. If your process is genuinely unusual, you will spend more time bending a skill than writing a prompt. The forkability claim is the escape hatch: the README states plainly that everything is MIT licensed so teams can review and adapt the workflows. Adapting is the expected path, not a workaround.
Maintenance, Licensing and What an Upgrade Actually Costs
The licence is MIT, which permits commercial use, modification and redistribution provided the copyright notice and permission notice are preserved. That is a statement about the licence text, not legal advice; if you redistribute a modified skill set inside a product, have your own counsel read the LICENSE file. The last push to the default branch was 2026-09-09, and the repository is not archived, so the project is active as of that date. Maintenance cost for an adopter has two parts. The cheap part is the skills: markdown files that you can pin, fork or edit, with no build step described in the README. The expensive part is the hosted NotFair MCP connection, because that is where authentication, account permissions and API surface changes land, and none of that lives in your repository. Upgrading the Claude Code plugin or refreshing the Codex marketplace re-pulls the skills, but it does not change what the MCP server is allowed to touch. Budget for a periodic re-read of whichever SKILL.md files you actually invoke, since a skill whose required inputs changed will fail in a way that looks like an agent mistake rather than a version mismatch.
Editorial conclusion
Adopt the NotFair Plugin if you already run Claude Code, Cursor, Gemini CLI or Codex and want auditable marketing procedures rather than a single vague prompt: the skills are plain SKILL.md files, the licence is MIT, and the read-only-first rule is visible in the repository. Do not adopt it expecting unattended campaign management, since the README describes approval boundaries around paid-media changes and the mutation surface is limited to what the connected accounts expose. Before installing, verify three things: that the NotFair MCP OAuth flow completes in your host, that your GA4, Search Console and ad accounts are readable by the connected user, and that you have read the specific SKILL.md for the workflow you intend to run, because that file is the contract for what the agent will and will not do.
Community notes