GEO Corpus Feeder

Problem

GEO Corpus Feeder targets brand visibility in generative AI engines such as DeepSeek, Kimi, and Doubao. It handles fragmented content, inconsistent structure, contradictory facts, and one-off optimization, turning brand, product, FAQ, case, and industry notes into a maintainable corpus managed by a monthly loop.

How It Works

On first run, it creates corpus/, schedule/, and templates/, then advances weekly:

• Week 1: scan corpus files and flag duplicates, stale items, and conflicts.
• Week 2: add YAML metadata, propose FAQPage, Product, or similar schema annotations, and output a logic conflict report.
• Week 3: split long passages, replace vague references, add contexts, generate FAQ pairs, and score GEO friendliness.
• Week 4: record corpus items, platforms, and status, establish a measurement baseline, and archive the month.

It supports commands such as 查看GEO进度, GEO月度报告, and 检查语料质量, and can use GEO_CORPUS_DIR to adjust paths.

Boundaries

The skill focuses on corpus organization, optimization guidance, and process records. It does not publish directly to platforms or guarantee model rankings. File edits, schema insertions, and feeding actions usually require user confirmation, and effect tracking depends on later monitoring data. Percentages in the source are better treated as optimization heuristics than reproducible benchmark results.