Skip to main content

openpost-seo

Audit OpenPost marketing and documentation SEO with the pinned claude-seo toolkit. Use for site audits, search discovery, structured data, sitemaps, or agent access checks.

معلومات المصدر

المستودع
getopenpost/openpost
آخر نشاط في المصدر
٢٦ سبتمبر ٢٠٢٦ في ١٨:٥٨
لغة SKILL.md المكتشفة
الإنجليزية
النجوم
٦٥٣
التفرعات
٨٢

خيارات التثبيت

يُحدَّد Prompt الذي يراجع المصدر أولًا بشكل افتراضي. يمكنك التبديل إلى أمر مباشر أو تنزيل نسخة محلية.

مراجعة ملفات المصدر

اقرأ SKILL.md وأي ملفات مرافقة يعرضها SkillsMP قبل أن تقرر التثبيت.

عرض SKILL.md

SKILL.md
تعليمات المصدر · معاينة للقراءة فقط
name
openpost-seo
description
Audit OpenPost marketing and documentation SEO with the pinned claude-seo toolkit. Use for site audits, search discovery, structured data, sitemaps, or agent access checks.
# OpenPost SEO audit Run the [claude-seo](https://github.com/AgriciDaniel/claude-seo) scripts directly from Codex. Its Claude Code slash commands and agents are reference material for choosing checks, not part of this workflow. ## Toolkit - Use the shallow, Git-ignored checkout at `docs/references/claude-seo`, pinned to tag `v2.4.0` (`e77e783e38eeb738424eb72117abbd2dacdd88af`). Clone that tag there if the checkout is absent. Review upstream changes before updating the pin. - Read `skills/seo/SKILL.md` and only the relevant leaf skill in that checkout for the requested audit. Execute bundled Python scripts through `scripts/claude-seo run <script.py>`, not a bare interpreter. - Set `CLAUDE_SEO_PYTHON` to a Python 3.10+ executable and `CLAUDE_SEO_DATA_DIR` to a dedicated user-data directory outside the repository. Run `scripts/claude-seo doctor --json` first. If it is not ready, run `scripts/claude-seo setup` only when the user requested setup or dependency repair, then confirm the doctor reports `ready: true`. The checkout's `install.sh` installs Claude Code skills globally and is unnecessary here. ## OpenPost audit 1. Identify the live target and the local Git revision. A local route absent from production may await release. Keep that distinction in every finding. 2. Check the live marketing and documentation sitemaps with `sitemap_discovery.py`. Use `parse_html.py` for titles, descriptions, headings, canonicals, social metadata, and JSON-LD. Use `agentic_check.py` for robots, no-JavaScript content, Markdown alternates, and agent access. The upstream DNS-pinned fetcher is single-threaded; run its requests serially. 3. Verify each reported issue against the live page and its source owner. Marketing metadata and routes come from `packages/social-images/src/index.js`, structured data from `apps/marketing/src/routes/_structured-data.ts`, sitemaps from `apps/marketing/src/routes/sitemap.xml/` and `apps/docs/app/sitemap.ts`, and agent-readable output from `scripts/generate-agent-surfaces.mjs`. Treat informational and not-applicable checks as such. A tool score alone does not prove search visibility or a defect. 4. Report concrete evidence, affected URLs, source revision or deployment state, and the next action. For code changes, follow `AGENTS.md` checks for the affected surface. Keep point-in-time audit output outside tracked source files. PageSpeed Insights may require credentials or hit a rate limit. Report the missing measurement and use another available source when performance is in scope; do not infer Core Web Vitals from HTML alone. Google treats `https://openpo.st` and `https://openpo.st/` as the same root URL. Normalize that pair when comparing sitemap and canonical values; see [Google's trailing-slash guidance](https://developers.google.com/search/blog/2010/04/to-slash-or-not-to-slash).
عرض على GitHub