news

News — News aggregation skill

Fetch the latest content from configured sources and output concise news summaries.This skill does not send HTTP requests directly, all network operations are delegated to search-and-fetch for execution.

Source configuration

All sources are defined in config/sources.yaml. Read that file to learn the currently available topics, direct URL sources, and daily briefing default topics.

For fetch policy details, see references/fetch-strategy.md, and for output format templates, see references/output-template.md。

Reading depth

This skill supports two reading depths, determined during intent parsing in Step 1:

Depth	Trigger signal	Behavior
Shallow read (default)	No special modifiers	Only fetch the list page, output title + summary + link, and prompt at the end that a deep read is available
On-demand deep read	After viewing the list, the user says "read item 2", "deep read this", "expand the first one"	Fetch the detail page for the specified item and output the full content
Automatic deep read	"deep read", "full text", "detailed content", "read all", "expand all"	After fetching the list, automatically fetch the first N detail pages (default N=3), executed in parallel

The user says "help me look at the detailed content of today's agents-radar issues" → automatic deep read The user says "give me a Hacker News briefing" → shallow read, with a prompt at the end

Workflow

Step 1 — Intent parsing

Based on the user input, simultaneously determinesourceandreading depth：

Source determination：

User intent	Basis for determination	Handling method
Daily briefing	"today's news", "morning report", "evening report", "briefing", no specific topic	Use the topics configured in `daily_briefing_topics` topics configured in
Specified topic	Mentions a specific field (AI, cryptocurrency, new energy...)	Match the corresponding key in `topics` corresponding key in
Specified source	Mentions a specific website/media name	Match from `direct_sources` match in
Custom query	The user provides specific keywords but they do not match any configuration	Directly call Search Mode with the user's keywords

If the user's topic is not in the configuration, do not report an error—just search directly with the user's keywords; this is a natural fallback.

Step 2 — Level 1: Fetch the list

Based on the parsing result, inthe same response round, call in parallel all applicable fetch actions:

Search-type topics(from the topics configuration):

For each search_queries entry, call Search Mode
If websites list is non-empty, append to the query site:domain.com limit source
multiple queries run in parallel without waiting

Direct URL category(from direct_sources configuration):

call Fetch Mode to scrape the list page, and receive it using integrated mode
extract each entry'stitle + URL + one-sentence description, take at most 10 entries

Parallelism principle: at most 3-4 parallel fetches each time; if exceeded, execute in rounds.

Step 3 — Level 2: Fetch details (depending on depth mode)

Shallow-read mode: skip this step and go directly to Step 4 output.

On-demand deep read: after the user sees the list and specifies entries → call Fetch Mode on the specified URLs, usingstandalone modeto output the full content (including five-dimensional analysis).

Automatic deep read: after list scraping is complete, immediately fetch the detail pages of the first N entries in parallel (default N=3):

call Fetch Mode on N URLs in parallel, integrated mode (title + summary + key points, skipping five-dimensional analysis)
N can be specified by the user ("deep read first 5"), otherwise default is 3
after detail fetching is complete, merge into the output, expanding each list item into a full content block

see references/fetch-strategy.md。

Step 4 — Consolidated output

Output in references/output-template.md format:

Group by topic/source
deduplicate the same URL, keeping the entry with the most complete content
timeliness first: entries from the last 24 hours are placed first
Shallow-read ending: always append 输入序号或标题可深读全文，或说"深读前N条"自动展开

Delegation protocol

When calling search-and-fetch:

list scraping / automatic deep-read details → integrated mode (title + summary + key points, skipping five-dimensional analysis)
on-demand deep read of a single article → standalone mode (full five-dimensional analysis)

Clearly state the mode when delegating, to avoid search-and-fetch outputting extra content.

Extension notes

To add a new information source, just edit config/sources.yaml, no need to modify this file:

New topic → add a key in the topics block
New URL source → append one entry to the direct_sources list
Adjust the daily briefing → modify the daily_briefing_topics list