Use Awesome Finance Skills to Build a Market News Briefing Agent

Reader persona: a founder, operator, analyst, or finance-curious product team member who wants an AI agent to summarize market signals without manually checking ten tabs every morning.

Job to be done: safely try RKiding/Awesome-finance-skills, install one concrete finance Agent Skill, and turn it into a repeatable market news briefing workflow.

Finance is one of the easiest places to make an AI agent sound impressive and still be useless.

A vague prompt like “tell me what matters in markets today” usually produces a confident summary with unclear sources, no repeatable process, and no way to know whether the agent looked at current data.

A finance skill is useful only if it makes the workflow more explicit:

• what data source should be checked;
• what tool or API is being called;
• what output the operator should expect;
• what risks or permissions are involved;
• what the agent should never pretend to know.

That is why RKiding/Awesome-finance-skills is worth a practical first run.

The X signal that triggered this guide was a high-engagement post about Awesome Finance Skills, a collection of open-source finance Agent Skills for Claude Code, OpenClaw, Cursor, OpenCode, and similar agents.

The repository validated for this guide was:

https://github.com/RKiding/Awesome-finance-skills

At validation time, GitHub showed the repository as active, with more than 2,300 stars and a recent update on 2026-05-22.

What the repository includes

The project positions itself as a finance Agent Skill collection for:

• real-time financial news and trends;
• A-share, Hong Kong, and US stock data;
• sentiment analysis;
• market prediction workflows;
• signal tracking;
• finance logic visualization;
• professional report generation;
• web search and local RAG.

The discovered skills included:

alphaear-news
alphaear-stock
alphaear-sentiment
alphaear-predictor
alphaear-signal-tracker
alphaear-logic-visualizer
alphaear-reporter
alphaear-search
alphaear-deepear-lite
skill-creator

For a first run, do not install everything.

Start with alphaear-news, because a daily news brief is easier to validate and safer than jumping straight into prediction or trading-like workflows.

First-run install

The repository README recommends installing individual skills with npx skills.

For a focused first run:

npx skills add RKiding/Awesome-finance-skills@alphaear-news

If you want a non-interactive install in a throwaway workspace, use:

npx skills add RKiding/Awesome-finance-skills@alphaear-news --yes

In validation, the command found 10 skills and installed one selected skill:

Selected 1 skill: alphaear-news
Installed 1 skill: ./.agents/skills/alphaear-news

The installer also displayed a security summary:

Gen: Safe
Socket: 0 alerts
Snyk: Med Risk

Do not ignore that last line. “Med Risk” does not mean “do not use it,” but it does mean the operator should understand dependencies and data access before using it in a serious workflow.

What alphaear-news does

The installed SKILL.md describes alphaear-news as a skill for:

• hot finance news;
• unified trend reports;
• prediction-market data;
• multiple sources such as Weibo, Zhihu, WallstreetCN, 财联社, Snowball, Hacker News, and Polymarket.

The skill points the agent to Python tools:

scripts/news_tools.py

Main capabilities:

fetch_hot_news(source_id, count)
get_unified_trends(sources)
get_market_summary(limit)

Supported source IDs include:

cls
wallstreetcn
xueqiu
weibo
zhihu
baidu
toutiao
douyin
thepaper
36kr
ithome
v2ex
juejin
hackernews

The important operator point: this is not just a prompt template. It gives the agent named sources and named functions.

That makes the brief easier to audit.

Minimal usage example

After installation, ask your agent for a narrow output:

Use alphaear-news to create a morning market brief.
Check cls, wallstreetcn, xueqiu, and hackernews.
Return:
1. top 5 signals;
2. why each matters;
3. source list;
4. what needs human verification;
5. no investment recommendation.

A good output should look like this:

# Morning Market Brief

## 1. Macro / policy signal
- What happened
- Why it matters
- Source
- Verification needed

## 2. Sector signal
...

## Human checks before action
- Confirm source timestamps
- Check whether headlines are duplicated across sources
- Check market open/close context
- Do not treat this as financial advice

If the agent gives you a confident buy/sell call, the workflow is wrong.

The job of this first-run brief is to compress signals, not to make trades.

Validation notes from the first run

The npx skills path worked for installing a specific skill with --yes.

The installed skill included:

SKILL.md
references/sources.md
scripts/news_tools.py
scripts/database_manager.py
scripts/content_extractor.py
tests/test_news.py

The Python implementation depends on:

requests
loguru

The stock-related skills may require heavier dependencies such as:

pandas
akshare
yfinance

In the validation environment, Python package installation was not available, so the script could not be executed end-to-end there. Direct endpoint checks showed a useful operational caveat:

• the Polymarket API returned HTTP 200 with JSON;
• the NewsNow endpoint returned a Cloudflare 403 from this environment.

That does not make the skill useless. It means your production agent should report source failures clearly instead of silently inventing a brief.

Permission and risk notes

This workflow touches finance information, so set boundaries up front.

Recommended agent rules:

- Always cite sources and timestamps when available.
- Never invent missing data.
- Mark API failures visibly.
- Do not produce buy/sell recommendations.
- Separate facts, interpretation, and uncertainty.
- Ask for human confirmation before sending any external report.

Recommended data rules:

- Treat public news APIs as unreliable until checked.
- Cache raw source responses if you need auditability.
- Avoid putting private portfolio data into generic prompts.
- If portfolio data is used, keep it local or in an approved database.

Recommended output rule:

This is market research assistance, not financial advice.

That disclaimer is not decoration. It helps keep the agent in the right lane.

A practical operator workflow

Use the skill as a daily research step:

08:30 — gather market and tech headlines
08:35 — group signals by macro, sector, company, sentiment
08:40 — identify contradictions or missing context
08:45 — produce a brief for human review
08:50 — optionally turn the brief into a Slack, Telegram, or email draft

The agent should not auto-send the final brief unless you have explicitly approved that channel and audience.

For teams, add a simple review checklist:

- Are all claims sourced?
- Are stale headlines removed?
- Are duplicated headlines merged?
- Are speculation and facts separated?
- Is there any accidental investment advice?

Where ClawMama fits

If you do not want to wire this up from scratch, ClawMama can host a ready-to-use OpenClaw/Hermes agent for this kind of recurring workflow.

A good starter setup:

• install alphaear-news in the agent workspace;
• run the brief once per weekday morning;
• keep source failures visible;
• draft the report for human approval;
• send only after confirmation.

New users get $2 credits, and the hosted agent can use the latest ChatGPT model for the summary and editing layer.

The point is not to replace an analyst.

The point is to make the first 20 minutes of market scanning repeatable, sourced, and reviewable.