Hot Topic Curation

Hot Topic Curation

Use this skill when the task sounds like:

• “做一个 AI 日报 / 热门话题专项”
• “抓几个外部热点站聚合到站内”
• “维护热点源、更新摘要页、顺手路由到对应专题”
• “像 AIbase Daily 那样做一个长期维护的热点聚合页”

This is a project-local editorial + ingestion skill. Prefer repo conventions over generic scraping habits.

Core Outcome

Produce or maintain a lightweight hot-topic aggregation flow that does all of the following:

1. fetch trusted public source pages or feeds
2. normalize them into a reusable local snapshot
3. keep the hub page updated
4. route worthwhile findings into the right site hotspot or topic pages
5. preserve source links and avoid full-text copying

Workflow

1. Read the source policy first

Before adding a new source, read:

• references/source-policy.md
• references/route-hints.md

This skill is deliberately conservative:

• prefer RSS / feed endpoints first
• use HTML daily pages when the site exposes a stable public digest page
• keep only title/link/date/one-line note level data in the local snapshot
• never turn this into a full-content mirroring pipeline

2. Inspect the current source registry

The canonical source registry is:

• site/data/hot-topic-sources.json

It defines which sources are watched, how they are parsed, and which site routes they usually feed.

Do not hardcode a new source directly into prose if it should become a durable maintained source. Add it to the registry first.

3. Refresh the local snapshot

Use the repo tool:

python3 tools/fetch_hot_topic_sources.py --write

Useful variants:

python3 tools/fetch_hot_topic_sources.py --limit 6 --json
python3 tools/fetch_hot_topic_sources.py --config site/data/hot-topic-sources.json --write

The default output snapshot is:

• site/data/hot-topic-snapshot.json

4. Keep the hub page aligned

The hot-topic hub is:

• site/topic-hot-watch.html
• site/md/topic-hot-watch.md

When the source registry or snapshot meaningfully changes, keep this hub page aligned. The page should explain:

• what kinds of sources are watched
• what is RSS-first vs HTML-digest fallback
• how the snapshot should be read
• which downstream site pages receive routed items

5. Route findings into the right topic

Do not keep everything trapped inside the hot-topic hub.

Use references/route-hints.md to decide where an item belongs:

• Agent / CLI / workflow news:
  • site/topic-agent-hot.html
  • site/topic-ai-cli-agent.html
• RAG / retrieval / evaluation:
  • site/topic-rag-hot.html
• model releases / API / pricing / infra:
  • site/topic-llm.html
  • site/topic-ai-api.html
• broader tool discovery:
  • site/topic-ai-zahuopu.html
  • site/topic-toolchain.html

If one concrete keyword deserves a proper draft, hand it off to:

• .claude/skills/keyword-site-topic/SKILL.md

6. Leave traceability

The hub and downstream pages should preserve:

• source site name
• original link
• date if available
• why it matters in one line

When summarizing, stay at digest level. This skill is for aggregation, routing, and maintenance, not article plagiarism.

Editing Targets

Default targets for this skill:

• site/data/hot-topic-sources.json
• site/data/hot-topic-snapshot.json
• site/topic-hot-watch.html
• site/md/topic-hot-watch.md
• hotspot pages such as site/topic-agent-hot.html or site/topic-rag-hot.html

Validation

After a maintained pass, prefer this sequence:

1. refresh the snapshot
2. update the hub or routed page
3. run:

python3 tools/refresh_site_topic_metadata.py
python3 tools/check_site_md_parity.py

If the loop-task board should reflect a new recurring line, also run:

python3 tools/build_loop_task_board.py