News Collect

Run a repo-local scan through skills/news-aggregator-skill, then normalize the result into a local report contract for the collect layer. Use when Codex needs a broad overseas or domestic media scan, wants a repeatable news-report.md, or needs raw JSON archived under content-production/inbox/raw/news/.

Use the repo-local news-aggregator-skill as a dependency, but always write outputs using this project's collect-layer contract. The report now includes lightweight写作决策字段 for recommended topics.

Quick Start

Run the default command:

bash

.venv/bin/python -m skill_runtime.cli run-skill news-collect --input content-production/inbox/20260405-ai-news-request.md

Prepare Input

Pass a markdown request file with YAML frontmatter.

Supported fields:

• profile: mixed_daily / global_latest / global_ai / cn_media / custom
• sources: only used when profile=custom
• keyword
• limit
• deep
• title

Example:

markdown

---
profile: mixed_daily
keyword: AI,Agent
limit: 8
deep: true
title: AI 每日资讯扫描
---

补充说明：更关注能延展成公众号观点文的题目。

Follow Collection Workflow

1. Read the request frontmatter and resolve the source profile.
2. Call skills/news-aggregator-skill/scripts/fetch_news.py with --no-save.
3. Save the raw JSON to the local inbox raw directory.
4. Rewrite the result into a stable markdown report for人工选题与后续深研。
5. For recommended items, also emit写作价值判断、推荐切口、推荐框架与标题方向。

Write Output

Write the report to:

text

content-production/inbox/YYYYMMDD-<slug>-news-report.md

Write the raw JSON to:

text

content-production/inbox/raw/news/YYYY-MM-DD/<slug>.json

Respect Constraints

• Only read the repo-local dependency under skills/news-aggregator-skill/
• Do not save outputs back into the vendor skill directory
• Keep the wrapper contract stable even if the vendor skill adds new sources later

Read Related Files

• Runtime entry: skill_runtime/engine.py
• Wrapper runtime: skills/news-collect/runtime.py
• Vendor dependency: skills/news-aggregator-skill/
• Data contract: docs/data-contracts.md