yuanyuanyuan430/ganhuo-geo-skill

Ganhuo GEO Engineer

What Problem Does This Skill Solve

ganhuo-geo-engineer specifically addresses a very common and often wasted type of content: old articles that have already been written, which indeed hold value, but suffer from loose structure, weak evidence, non-front-loaded answers, incomplete FAQs, and low direct extractability by AI.

Its goal is not "mystically guaranteeing AI citations," but rather to transform old content into assets that are easier for humans to read, for AI to summarize, for sales to reuse, and for knowledge bases to absorb.

The core principles are simple:

• Preserve the original meaning, do not alter viewpoints;
• Diagnose first, then rebuild;
• Write substantively when material exists, mark as "pending supplement" when it does not;
• Transform articles from "author monologue" to something where both readers and AI can quickly grasp answers;
• Finally deliver a complete, editable, publishable, and reusable package.

When to Use

Use this Skill when the user provides the following types of content:

• WeChat public account (official account) old articles;
• Official website articles, product pages, service pages;
• Tutorials, FAQs, knowledge base articles;
• Founder's notes, case studies, industry observations;
• Local files such as .md, .txt, .html, .docx, .pdf;
• A batch of historical content that needs unified upgrading.

If the user only provides a topic and wants to write a new article from scratch, do not use this Skill directly. First ask the user to provide old articles, or switch to a new content creation workflow.

Required Readings

• references/geo-playbook.md for the rebuild workflow.
• references/geo-output-standard.md for the delivery package.
• references/geo-anti-patterns.md when the request includes ranking promises, fake proof, keyword stuffing, or thin content.
• references/geo-material-notes.md when the user provides mixed materials or asks for evidence handling.
• templates/article-brief.md when the user needs an input template.

Execution Workflow

1. Read the input. If the article cannot be read, ask the user to paste the body text.
2. First preserve the original title, core viewpoints, facts, cases, tone, and constraints.
3. Audit the material: core arguments, evidence, data, quotes, cases, entities, terminology, structure, questions, reusable sentences, missing materials.
4. Score conservatively: evidence, structure, answerability, entity clarity, semantic coverage, credibility signals, terminology, stability, readability, keyword stuffing risk.
5. Select rebuild depth:
   • light: preserve original structure, supplement core answers, FAQ, and light cleanup;
   • standard: restructure, add scorecard, FAQ, pending supplement list, and risk warnings;
   • deep: rebuild into a complete content asset with sections, tables, FAQ, pending supplement list, and multi-channel reuse suggestions.
6. Rewrite by priority: front-load answers, support claims with evidence, create reusable paragraphs, FAQ, entity explanations, actionable steps, boundary constraints, readability.
7. Any new content not supported by the user's material must be placed in the pending supplement list. Do not fabricate data, clients, quotes, rankings, product capabilities, or research conclusions.
8. Return the complete delivery package according to references/geo-output-standard.md.
9. Finally conduct a quality check: whether the original meaning is preserved, whether there is fabricated evidence, whether keyword stuffing exists, whether the pending supplement list is clear, whether the structure is publishable.

Default Deliverables

1. Rewritten body text
2. GEO scorecard
3. Rebuild notes
4. Pending supplement list
5. Risk warnings

For scenarios such as AI tutorials for practical work, Feishu documents, or official website knowledge bases, also add:

6. Reuse suggestions for Feishu, official website, WeChat public account, Zhihu, sales FAQ, and knowledge base

Quality Baseline

Better to write less than to be vague.

Every strong conclusion must be in one of the following four states:

• Already supported by the user's material;
• Verified during task execution;
• Explicitly placed in the pending supplement list;
• Removed or weakened.

Never fabricate missing data as real data. Never promise rankings, traffic, citations, recommendations, or model preferences.

Pre-release Validation

Before releasing to the repository, run:

python3 scripts/validate_skill.py