🤖 Daily AI Newsletter (Brave + Firecrawl)
An OpenClaw skill that:
- Uses Brave Search (
web_search) to find recent AI‑related news.
- Filters and ranks results to keep ≈20 strong‑relevance items.
- Uses Firecrawl (
web_scrape) to scrape each chosen article.
- Summarizes each article with the model (
language_model).
- Produces a structured
newsletter_items list, plus markdown_newsletter and json_newsletter outputs for your newsletter engine.
🧩 Prerequisites
Before this skill runs, you must:
- Install OpenClaw (Docker, pip, or GitHub clone).
- Expose MCP‑compatible tools or configure them in
config.yaml.
- Register Brave Search API:
- Register Firecrawl:
- (Optional) Install and authorize Composio:
pip install composio_pydantic
composio login
composio add firecrawl
- Define two helper tools (or ensure they exist in your OpenClaw):
filter_ai_news – filters and ranks search results.extract_urls – extracts URLs from a list of articles.
🧭 Skill Overview
- Goal: Produce a daily AI‑focused newsletter with ≈20 articles.
- Method:
- Search latest AI news with Brave (
web_search).
- Filter & rank results (
filter_ai_news).
- Extract URLs of selected articles (
extract_urls).
- Scrape & enrich each URL with Firecrawl (
web_scrape) + summarization (language_model).
- Compile into
markdown_newsletter and json_newsletter for your engine.
🪄 Skill Definition
skill: ai_newsletter_daily
description: Daily AI‑focused newsletter via Brave + Firecrawl.
version: 1.0.0
inputs:
- name: target_news_count
type: integer
default: 20
description: Number of AI articles to include (must be ≥ 1).
- name: search_query
type: string
default: "latest AI news today"
description: Query to send to Brave Search.
- name: include_domains
type: list[string]
default: []
description: Optional whitelist of domains (e.g., "arxiv.org").
- name: exclude_domains
type: list[string]
default: ["youtube.com", "reddit.com", "facebook.com", "twitter.com"]
description: Domains to skip.
outputs:
- name: newsletter_items
type: list[object]
description: |
Up to `target_news_count` AI‑related articles, each with
title, URL, domain, summary, published_at, and relevance_score.
- name: markdown_newsletter
type: string
description: Markdown‑formatted newsletter body for your engine.
- name: json_newsletter
type: object
description: |
JSON‑structured newsletter with date and article list
for your email / RSS / CMS pipeline.
parameters:
- name: max_brave_results
type: integer
default: 50
description: Max number of results from Brave before filtering.
- name: filter_by_ai_keywords
type: boolean
default: true
description: Whether to score relevance by AI‑related keywords.
🧩 Step 1: Search for AI news (Brave)
Use web_search to get raw AI‑related links from Brave.
task:
name: search_ai_news
description: Fetch raw AI news results from Brave Search.
steps:
- call:
tool: web_search
args:
query: "{{ search_query }}"
count: "{{ max_brave_results }}"
- set:
var: raw_results
value: result.tool_output.results
# or `tool_output.items` depending on your Brave adapter shape
🧩 Step 2: Filter and rank results
Run a helper tool that filters and ranks the raw results.
task:
name: filter_ai_articles
description: Filter and rank AI‑related articles from raw results.
depends_on:
- search_ai_news
steps:
- call:
tool: filter_ai_news
args:
results: raw_results
targets: target_news_count
include_domains: include_domains
exclude_domains: exclude_domains
filter_by_ai_keywords: filter_by_ai_keywords
- set:
var: kept_articles
value: result.tool_output.filtered_results
🔧 filter_ai_news implementation (in your code, not in YAML) should:
- Remove duplicates by hostname+path.
- Filter by
include_domains/exclude_domains.
- Score relevance to AI (using LLM or keyword‑based scoring).
- Sort by relevance + freshness.
- Take only
targets items.
🧩 Step 3: Extract URLs
Turn filtered kept_articles into a plain URL list.
task:
name: extract_urls
description: Extract article URLs from filtered results.
depends_on:
- filter_ai_articles
steps:
- call:
tool: extract_urls
args:
articles: kept_articles
- set:
var: article_urls
value: result.tool_output.urls
🔧 extract_urls implementation:
- Should map
[{ title, url, ... }] → urls (i.e., urls = [article.url for article in articles]).
🧩 Step 4: Deep‑dive scrape with Firecrawl
For each URL, scrape with Firecrawl and summarize with the model.
task:
name: scrape_and_enrich_articles
description: Scrape and enrich each selected article.
depends_on:
- extract_urls
steps:
- for_each:
items: article_urls
item_var: article_url
steps:
- call:
tool: web_scrape
args:
url: "{{ article_url }}"
- set:
var: raw_content
value: result.tool_output
# Assumed to have metadata: { title, published_at, ... }
- call:
tool: language_model
args:
model: gpt‑4
prompt: |
You are a summarizer for an AI‑focused newsletter.
Given this article content from Firecrawl:
---
{{ raw_content }}
---
Summarize it in exactly one short paragraph.
Do not include code or technical details unless crucial.
Return only the paragraph, no headings or extra text.
- set:
var: summary
value: result.tool_output.text
- call:
tool: enrich_article
args:
article:
title: raw_content.metadata.title
url: article_url
domain: raw_content.metadata.url_domain || extract_domain(article_url)
summary: summary
published_at: raw_content.metadata.published_at || "unknown"
- set:
var: enriched_item
value: result.tool_output.article
- append:
list: enriched_articles
value: enriched_item
🔧 enrich_article implementation:
- Should fill in
domain from url and compute relevance_score (e.g., via keyword‑pattern or lightweight LLM scoring).
🧩 Step 5: Compile the newsletter (20 items)
Sort and truncate to target_news_count, then render Markdown and JSON.
task:
name: compile_newsletter
description: Turn enriched articles into structured newsletter output.
depends_on:
- scrape_and_enrich_articles
steps:
- sort:
list: enriched_articles
by: "relevance_score DESC, published_at DESC"
- set:
var: final_list
value: enriched_articles[0:target_news_count]
# or via `tool: limit_list` if your runtime prefers tool calls
- set:
var: markdown_output
value: |
# 🤖 Daily AI Newsletter — {{ today }}
{% for item in final_list %}
## {{ item.title }} ({{ item.domain }})
{{ item.summary }}
[Read more]({{ item.url }})
{% endfor %}
- set:
var: json_output
value: |
{
"date": "{{ today }}",
"articles": final_list
}
- output:
newsletter_items: final_list
markdown_newsletter: markdown_output
json_newsletter: json_output
🚀 How to deploy on ClawHub
-
Save this file as:
skills/ai_newsletter_daily.md
in your OpenClaw project or ClawHub skill folder.
-
Make sure your OpenClaw stack exposes these tools:
web_search → Brave Search.web_scrape → Firecrawl.language_model → your LLM provider.filter_ai_news, extract_urls, enrich_article → helper tools implemented in your codebase.
-
In ClawHub, trigger the skill:
- As a daily cron job (e.g.,
0 9 * * *).
- Or via a webhook that overrides
search_query and target_news_count.
-
Consume outputs:
markdown_newsletter → send to your Markdown‑based newsletter engine.json_newsletter → pipe into your email engine, RSS generator, or CMS.
✅ Expected behavior
- When
ai_newsletter_daily runs:
- Brave returns ≈50 AI‑related links.
filter_ai_news keeps ≈20 high‑relevance items.web_scrape + language_model enrich each article.compile_newsletter yields:
newsletter_items – structured list of 20 AI‑focused items.markdown_newsletter – ready‑to‑render Markdown.json_newsletter – JSON‑endpoint‑ready payload.
You can now land this as a production‑ready ClawHub‑style skill and extend filter_ai_news, enrich_article, etc., in your OpenClaw code.
