한국어 · English

Paying $400–$2,500/month for an SEO or "AI visibility" agency? Fire them. Your AI agent can do the work.

A wave of services now charges monthly retainers for "AI-era search optimization", "guaranteed ChatGPT citations", and "LLM visibility". Most of what they actually do is public standards and repeatable checklists. When a human does it, it's a retainer. When your agent does it, it's a skill.

This repository is a Claude Code skill. Install it and your agent audits your site, fixes what's broken, and measures the results.

This is not theory. The same playbook, applied to Chickenstock — a solo-built Korean stock research service:

• 1.54M search impressions in 30 days (+85,578% MoM), 7.4K clicks
• Pages cited paragraph-by-paragraph by Naver's AI Briefing (Korea's answer engine)
• Zero ad spend — all inbound from search and AI citations

The journey is documented publicly on Threads @kindainvestor.

"Search optimization" means different things to different engines. This skill treats them as five distinct lanes:

NEO is what makes this skill different. Global AEO guides ignore Naver entirely, but if you serve the Korean market, half your traffic lives there.

Plus a content engine. Most of a retainer pays for "N posts a month". This skill replaces that with a pipeline: a question backlog pulled from your own search data, a frontmatter content model that generates the JSON-LD, a publish gate every post must pass, and a refresh/merge policy so posts don't rot. Output is measured as "posts that get seen and cited", not "posts published".

As a plugin (recommended — one command, easy updates):

/plugin marketplace add leopard627/fire-your-seo-agency
/plugin install fire-your-seo-agency@fire-your-seo-agency

Or via git clone:

# As a project skill (this project only)
git clone https://github.com/leopard627/fire-your-seo-agency.git .claude/skills/fire-your-seo-agency

# Or as a personal skill (every project)
git clone https://github.com/leopard627/fire-your-seo-agency.git ~/.claude/skills/fire-your-seo-agency

Then in Claude Code:

/fire-your-seo-agency audit my site

Your agent starts with a crawler-eye audit and returns a scorecard like this before touching anything:

…then proposes priorities, implements them, and schedules the re-measurement.

1. Audit — reads your site the way a crawler does (no JavaScript) and scores all five lanes
2. Technical base — fixes SSR exposure, sitemaps, meta, structured data
3. Intent landing pages — designs pages on the "one question = one page" principle
4. Machine readability — llms.txt, JSON-LD, citation-ready paragraph structure
5. Naver — from Search Advisor registration to AI Briefing citation requirements
6. Content engine — sub-blog setup, question backlog, publish gate, refresh & merge policy
7. Measurement loop — schedules a re-measurement and proves the change with numbers

• ❌ Buying backlinks, engagement pods, content spam — we don't fight the search engine
• ❌ "Guaranteed rankings" promises — no claims without measurement
• ❌ Keyword stuffing, hidden text, cloaking — nothing that gets your domain killed

The whole philosophy fits in one line: AI doesn't cite good writing. AI cites accurate data. Become the primary source for a number, and the citations follow.

SKILL.md              ← agent operating procedure (audit → implement → measure)
references/
  seo.md              ← technical SEO checklist + real-world traps
  aeo.md              ← answer engine optimization (Bing WMT · AI Overviews · Copilot · E-E-A-T)
  geo.md              ← generative engine optimization (AI crawler policy · llms.txt · primary source)
  llmo.md             ← model-knowledge optimization (brand entity)
  neo-naver.md        ← Naver (Search Advisor · AI Briefing · blog two-track)
  content.md          ← content ops (sub-blog · content model · backlog · publish gate · refresh)
  measure.md          ← the measurement loop (fixing it is not the finish line)
  en/                 ← English mirrors of all reference docs (for human readers)
.claude-plugin/       ← plugin & marketplace manifests (/plugin install support)
tools/                ← repository checks (validate.py) + lossless asset optimizer
tests/                ← unit tests for the checks
.github/workflows/    ← CI: checks, tests and the asset check on every push/PR

The Korean documents under references/ are canonical (the agent reads those); references/en/ mirrors them in English for human readers.

This repo checks itself — .github/workflows/validate.yml runs all of it on every push and pull request:

python3 tools/validate.py --strict          # 10 checks, standard library only
python3 -m unittest discover -s tests -t .  # unit tests for every check
python3 -m pip install pillow
python3 tools/optimize_assets.py --check    # flags recompressible assets

tools/validate.py enforces the things that rot silently in a docs-only repo: the SKILL.md frontmatter contract (kebab-case name, 64-character name limit, 1024-character single-line description), plugin.json ↔ marketplace.json version and name agreement, CHANGELOG.md matching the manifest version, relative links and anchors, Korean/English reference parity (heading counts, table rows, checkboxes, section numbering), the asset budget and preview dimensions, the MIT license metadata, and a secret scan over every text file.

MIT — use it freely, and keep the retainer money.