메뉴
Research and maintain an evidence-backed SEO/GEO thesis, opportunity portfolio, and execution-ready roadmap.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Convert an approved SEO/GEO strategy into independently executable tracker work packages.
Deliver one tracker work package labeled `ready-for-seo-agent` through acceptance → delivery → evidence.
Codebase-driven SEO + GEO (Generative Engine Optimization) audit. Gathers site facts into SEO.md, scores the site against a fixed rubric, and writes a dated report with a prioritized improvement plan.
Turn an ADO user story — or a rough idea — into a verifiable, code-grounded blueprint a TDD executor can build from. Use when the user wants work planned before coding.
Create a PRD with an embedded Design Direction section through user interview, codebase exploration, and design intent capture. Use when building or revamping any feature with frontend or UI work — landing pages, components, UI redesigns — where design intent needs to be locked in before implementation. Extends the standard PRD process with structured design thinking: audience, tone, aesthetic direction, and the one memorable differentiator.
Right-sized guided feature development with three presets (Quick / Standard / Deep) covering everything from one-line tweaks to risky new modules. Invoke for any non-trivial feature work.
| name | seo-strategy |
| description | Research and maintain an evidence-backed SEO/GEO thesis, opportunity portfolio, and execution-ready roadmap. |
| disable-model-invocation | true |
| argument-hint | Optional: capture an idea, validate a keyword/topic, discover opportunities, compare directions, or refresh the active strategy |
Decide where a site should compete next. Form a defensible thesis, select one primary and one secondary strategic bet, and document a six-month direction with a 90-day initiative roadmap. Do not implement the roadmap.
The skill may read an existing SEO.md or seo-reports/, but an audit is optional: never invoke /seo or duplicate its rubric automatically. Apart from the artifacts below, remain read-only.
On a bare invocation, use Refresh when SEO-STRATEGY.md has an active strategy; otherwise use Discover. Infer context before asking; accept unknown and disclose a safe assumption when the user cannot supply a fact.
For Capture, write a Captured proposal using references/templates.md. Do not perform live research, form a thesis, or alter the active strategy.
Read the current strategy artifacts, relevant repository files, and live site when known. Establish the branch, business outcome, valuable conversion, audience, primary market, language, and decision being made. Treat user-supplied capacity as optional.
The outcome, conversion, and audience are the user's facts, not the agent's. Present the inferred frame — what was found, what was assumed, what stayed unknown — and have the user confirm or correct it before research is spent on it. When the user is unavailable, proceed on the disclosed frame and keep the eventual thesis Proposed.
Complete when: the user has confirmed or corrected the frame, or an unattended run has disclosed every assumption in it.
Follow references/research-method.md for the selected branch. Write a dated research snapshot from references/templates.md. Live external research is required unless the user explicitly asks for a preliminary offline pass.
Complete when: material claims are sourced and dated, unavailable metrics remain unknown, and the branch's research frontier is covered.
Apply references/opportunity-framework.md. Evaluate every credible option or record why it was screened out. Use qualitative judgments, not an aggregate numeric score.
Complete when: the portfolio exposes the meaningful trade-offs and no plausible direction has disappeared without a reason.
Present the portfolio's leading options and the trade-offs between them, name a recommended primary and secondary bet, and get the user's steer before writing the thesis: appetite, capacity, and plans are their call, and the evidence cannot see them. When the user is unavailable, record the recommendation as the steer and keep the thesis Proposed.
Then state why the chosen bets beat the alternatives, with separate SEO and GEO rationales, risks, dependencies, and a lightweight measurement contract for each.
Complete when: the user has steered the choice (or an unattended run recorded the default), and the recommendation connects the business outcome to the evidence, bets, and continue/adjust/stop rules.
Present the thesis and ask the user to approve or revise it. Never silently replace an active strategy. If approval is unavailable, preserve the proposal according to references/opportunity-framework.md.
Complete when: the thesis is approved, revised, rejected, deferred, or explicitly left proposed.
Use references/templates.md to write the current strategy, proposal state, and any required history snapshot. The roadmap contains initiatives, not tickets: each initiative names its outcome, scope, deliverables, dependencies, completion evidence, measurement link, capabilities, phase, and review point.
Complete when: the artifacts link to one another, the current lifecycle state is unambiguous, the 90-day roadmap is execution-ready, manually maintained notes are preserved verbatim, and no implementation work was performed.
