사용 방법
- 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
- 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
- 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/sergekostenchuk/ui-ux-agent-skill-system --skill llm-citation-monitor시작 전 확인
- 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
- 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 Skill 문서
원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.
LLM Citation Monitor
Use this skill after the target site's SEO/LLM discovery layer exists and the user wants to measure assistant/search citation visibility.
Read references/citation-evidence-policy.md before producing a monitoring plan or report.
Owns
- target question matrix;
- manual observation workflow;
- citation report template;
- assistant/search surface evidence policy;
- competitor citation capture;
- caveats for personalization, locale, freshness, and account state;
- refusal of unsupported citation claims.
Does Not Own
- assistant account credentials;
- bypassing bot protections;
- scraping assistant/search surfaces without approval;
- changing robots, WAF, auth, or crawler policy;
- ranking claims;
- content/schema implementation.
Workflow
- Define target site, page set, audience, locale, and target questions.
- Classify query intent: brand, topic, problem, comparison, how-to, or citation-check.
- Choose approved observation mode: manual, exported transcript, screenshot, API/tool report, or public fetch.
- For each run, record surface, model/search product if visible, date/time, locale, account state, query text, answer summary, cited URLs, quote/snippet, and caveats.
- Capture competitor citations separately from target citations.
- Mark missing citations as
not observed, not as "does not cite anywhere". - Produce a report using assets/citation-report.template.md.
- Hand off site-content or technical fixes to SEO/LLM architecture skills.
Non-Negotiables
- Do not claim "ChatGPT cites us" without direct citation evidence.
- Do not treat one personalized answer as universal ranking.
- Do not store assistant credentials, cookies, account tokens, or private conversation data in skill files.
- Do not bypass paywalls, auth, robots, WAF, or tool restrictions.
- Do not scrape surfaces against terms of service.
- Do not publish screenshots/transcripts containing private user data.
Safety And Privacy Boundaries
- Reports should contain only approved public query text, public cited URLs, and redacted/private-safe snippets.
- If a screenshot or transcript includes account details, remove them before adding to shared artifacts.
- If a user provides a private assistant transcript, summarize it locally and ask before storing it.
- Treat model version, locale, date/time, and account state as part of the evidence, not incidental metadata.
Evidence Labels
Cited: target URL appears as a cited/source URL in the answer or citation panel.Mentioned: target brand/page mentioned but not cited.Not observed: no target citation in this run.Competitor cited: another domain was cited for the target question.Invalid evidence: missing timestamp, surface, query, or cited URL.
Validation
Validate skill edits with:
bash
python3 $CODEX_HOME/skills/senior-skill-architect/scripts/lint_production_skill.py $HOME/SKILL/skills/llm-citation-monitor
python3 $HOME/SKILL/plans/seo-llm-skill-cluster/scripts/lint_skill_cluster.py $HOME/SKILLForward tests live in evals.json.
Output Shape
Return:
- Monitoring objective and target pages.
- Query matrix.
- Approved surfaces and observation mode.
- Citation evidence table.
- Competitor citation table.
- Caveats and limitations.
- Fix backlog routed to the correct skill owner.
