CommunityWriting & Editinggithub.com

flaqai/backlink_skills

Orchestrate blog tasks: confirm requirements, assign authors and independent reviewers, accept complete HTML/Markdown full text with public image URLs, and check the published page. Used to start, resume, or deliver a blog project.

What is backlink_skills?

This skill orchestrates end-to-end blog production workflows. It acts as a central controller (G) that confirms requirements from human input, assigns writing tasks to authors (W) and independent reviewers (R), verifies final deliverables, and checks the published page. The final output must consist of two complete full-text versions with images: an HTML version with public image URLs inlined, and a Markdown version referencing the same URLs in the same positions. The skill ensures quality and efficiency by following a strict sequence: gather brand, style, keyword, CTA, and disclosure requirements; compile research and writing proposals per article/platform; obtain user confirmation before writing; assign limited visible authors and reviewers; validate deliverables against original requirements; and perform a public page QA check after publication. It supports any blog platform, enforces image hosting via public URLs, and requires all publishing actions to be done manually by a human. The skill is designed for agent platforms like Claude Code, Cursor, or Codex that can orchestrate sub-agents and manage file-based deliverables.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/flaqai/backlink_skills/tree/HEAD/scalable-seo-blog-workflow/skills/blog-gate

Installed? Explore more Writing & Editing skills: steipete/notion, langchain-ai/langchain, bytedance/podcast-generation · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Blog Master Controller

G is responsible for confirming requirements, assigning W (author) and R (independent reviewer), accepting deliverables, and checking the published page. Quality and efficiency are equally important.

The final delivery consists of two complete full-text articles with images: HTML displays public URL images in the body at the correct position, and Markdown references the same URL at the same position. The body contains only reader-facing content and captions; instructions, file names, visible alt text, etc., are listed separately outside the body. Pages containing only a list of image links, summaries, or placeholder cards are considered incomplete deliveries.

Work Order

  1. Collect brand and style, number of articles, candidate platforms, language/market, keywords, CTA (call-to-action links), disclosures, images, and delivery requirements. For each new task, provide a necessary manual preparation checklist, and ask for any missing items that affect delivery.
  2. Summarize research and writing proposals per article. Platforms are limited to the user-provided range, one article per platform; first read the overall content of the target column, then read articles from different sources to avoid misjudgment based on a single example. Platform materials help determine language, readership, and depth; the model designs topics accordingly. For multiple platforms, clarify reader questions and main help separately; localize the same topic according to user requirements. Deliver recommended topics, body arrangement, steps taken by the product, and brief justifications for confirmation before writing. Adjust the plan when unfavorable evidence is found; do not require existing popular articles on the same topic. Only fill gaps that would affect recommendations or key facts; stop collecting when enough material is available. Treat general marketing restrictions as risk warnings, keep candidates, and let the user decide trade-offs; continue with confirmed local tasks. See Platform Matching for matching methods; reusable read-only research sub-agents can be used.
  3. Present the complete plan, and after obtaining user confirmation, arrange writing and image generation. Save shared requirements, per-article reader tasks, market, differentiation angle, factual boundaries, and delivery arrangements in a single task-plan.json, preserving the original confirmation source. Reuse applicable platform materials for the same batch; what is confirmed are requirements and reader tasks; tentative titles, outlines, and examples are provided for authors to optimize independently, except for explicitly locked items.
  4. Reuse a small number of visible W and independent R. Each W writes in articles/<ID>/, G maintains shared requirements and progress. Provide role-specific requirements and applicable materials for each role; changes to scope, platform, language, CTA, or reader tasks are decided by the user; questions only pause the affected article.
  5. W completes and checks the final HTML/Markdown payload and assets; R reviews. After R passes the current version, G checks against the user's original confirmation: count, platform/language, keyword meaning, CTA/disclosure, link conventions, body images, additional cover image, and dual payload entry, then formally delivers to the user. Send specific quality issues back to the original W for fixing and the original R for re-review; report missing review results or unresolved major issues first.
  6. After manual publication, return the URL and an acceptance/needs-fix conclusion. The same G performs a default check of the desktop published page according to Public Page QA; judge based on actual rendering whether images break the layout, are abnormally cropped, or affect readability; check the mobile view only if the user requests it. Provide platform-side issues as merge/fix hints for a human to resolve; G only rechecks related items, does not reopen writing or review. Save progress and end execution when waiting for user messages.

Requirements & Tools

  • The visible anchor text of any outbound link to the target product must contain the corresponding confirmed brand name, preferably naturally describing its purpose. Determine the brand writing before writing begins; if the locked copy does not include the brand, ask the user to resolve it, do not automatically change the text. Specified CTAs, counts, and disclosures must be preserved; recommended CTAs should be relevant and secondary. The default relationship is "founder"; account operations require additional authorization.
  • Each title/expected H1 should naturally express all protected meanings of the keyword; when word-by-word locking is required, follow the original. For cross-language expression, choose in order: the brand's existing phrasing, regional natural variants, and documented translations.
  • Images should echo the brand as much as possible; default to providing an additional cover image beyond body images. Quantity and format follow confirmed values; view each final image. For new tasks, W selects the image host, creates public direct links, and delivers the complete HTML/Markdown pages with images; the plan includes image publicity requirements.
  • New tasks use the Lightweight Delivery Tool; old campaign.json projects follow the Old Interface to maintain original confirmations and records.
  • Blog publishing is performed only by a human: the machine must not input into the article editor, upload images to the blog platform, save drafts, publish, delete, or modify the fingerprint browser client. Image uploads are limited to publicly sharable images for the current article, following Image Hosting; registration/login and minimum setup follow Special Authorization, preserving existing model configurations.

After compression or recovery, fully read Short Workflow, then re-read the original task, current draft, and open issues; read referenced materials as needed.

Individual skills in this repo

This repo contains 1 individual skill — each has its own dedicated page.

Related Skills