geo-content-writer（地理内容作家）

Content Writer

Use this skill to turn Dageno prompt opportunities into a real-fanout backlog and then produce one backlog-row-first editorial package for one selected fanout item.

Fixed Workflow

A. Opportunity Layer

1. discover high-value prompts
2. extract real fanout for each prompt
3. store all fanout in one backlog

B. Backlog Layer

1. mark overlap / merge / duplicate items
2. keep one prioritized backlog with statuses
3. choose which fanout item to write next

C. Writing Layer

1. crawl top citation pages for the selected fanout
2. analyze citation patterns
3. build one editorial brief from one selected backlog row
4. generate section drafting instructions
5. generate section review instructions
6. assemble one publish-ready article

D. Distribution Layer

1. publish to WordPress draft or publish status

Input -> Output Contract

Inputs

• required: DAGENO_API_KEY
• required: one date window (days) for opportunity discovery
• optional: knowledge/brand/brand-knowledge-base.json
• optional: one explicit backlog_id
• optional: existing backlog file path

Outputs

• fanout backlog JSON (real fanout first; optional exploratory fallback rows are explicitly tagged)
• one publish-ready payload JSON (editorial_brief, draft_package, review_package)
• one decision-grade markdown article
• optional WordPress draft/publish handoff

Non-Negotiable Rules

• only use real Dageno fanout
• do not generate guessed fanout as publish-ready seed
• exploratory fallback is allowed only when write_now inventory is low, and must stay status=exploratory until validated against fresh GEO data
• do not write directly from Dageno topic
• one selected fanout should map to one article
• one backlog row should map to one editorial brief
• use the section drafting and review contracts when integrating with external agents
• if local brand knowledge base and Dageno brand snapshot do not match, block publish-ready output

- you can override with --allow-brand-mismatch, but it will carry a warning; avoid unless you intentionally accept risk

Output Quality Contract (Required)

• include explicit exclusion boundaries (not ideal when ...) for major options
• include a default recommendation hierarchy (forced ranking fallback)
• include at least one head-to-head comparison block between major options
• include an If X -> Choose Y decision engine section
• include a single-sentence convergence block (If You Only Remember One Thing)
• include at least 5 references with a mix of editorial and official support/policy pages

Quality gate command:

PYTHONPATH=src python -m geo_content_writer.cli check-article-quality <article.md> --min-words 1200

Required Local Files

• knowledge/brand/brand-knowledge-base.json
• knowledge/backlog/fanout-backlog.json

Reference

See references/pipeline-spec.md.