文件内容
SKILL.md
---
name: ai-news-radar
description: Fetch curated AI news, social signals, blogs, papers, events, and skills
from the Agentic Brew public RSS feeds (https://www.agenticbrew.ai/feed/*.xml) and
return a compact, agent-friendly list. Use when the user wants "today's AI news",
"what's trending in AI", "AI papers this week", "AI events", "AI blogs", "trending
AI repos", "trending AI on Reddit / YouTube / Product Hunt", "trending AI skills",
"ai news radar", "agentic brew feed", "AI signal radar", "latest AI digest",
or any request to pull curated AI items from Agentic Brew without writing a scraper.
---
# Agentic Brew Feed Fetcher
Pulls items from the Agentic Brew public RSS endpoints and returns them as a clean list. No auth, no scraping — just an HTTP GET against the latest run_log's published feed.
## Available feeds
| Feed | URL | Contents | Item `<link>` resolves to |
|----------------|--------------------------------------------------|----------|---------------------------|
| `news` | https://www.agenticbrew.ai/feed/news.xml | Synthesized news clusters — title + overview | Agentic Brew news-analysis card page (`https://www.agenticbrew.ai/news#cluster=<id>`) |
| `twitter` | https://www.agenticbrew.ai/feed/twitter.xml | Trending X / Twitter topics — title + hottest tweets with likes / RTs / replies / views | The top tweet of the topic on x.com |
| `github` | https://www.agenticbrew.ai/feed/github.xml | Trending GitHub AI repos — title + detail (stars, language, daily delta) | Original GitHub repo |
| `reddit` | https://www.agenticbrew.ai/feed/reddit.xml | Trending Reddit AI threads — title + detail (subreddit, upvotes, comments, excerpt) | Original Reddit thread |
| `youtube` | https://www.agenticbrew.ai/feed/youtube.xml | Curated AI videos — title + summary | Original YouTube video |
| `product_hunt` | https://www.agenticbrew.ai/feed/product_hunt.xml | Trending AI launches — title + topics + tagline | Original Product Hunt launch page |
| `skill` | https://www.agenticbrew.ai/feed/skill.xml | Top Claude Code skills from skills.sh + clawhub — title + installs/stars + summary | Original skill page on skills.sh / clawhub |
| `blog` | https://www.agenticbrew.ai/feed/blog.xml | Curated AI blog articles — title + AI-generated summary | Original blog article |
| `paper` | https://www.agenticbrew.ai/feed/paper.xml | Research papers — title + AI summary + institutions + source (HF/AlphaXiv/X) + votes | Original paper page (arxiv / Hugging Face / x.com) |
| `event` | https://www.agenticbrew.ai/feed/event.xml | Upcoming AI events — title + start time + summary | Original event page (e.g., lu.ma) |
| `all` | https://www.agenticbrew.ai/feed/all.xml | Union of all of the above | Per-item — same as the feed above |
## Usage
```
/ai-news-radar [feed] [--limit N] [--query KEYWORD] [--json]
```
- `feed` (optional, default `news`): one of `news`, `twitter`, `github`, `reddit`, `youtube`, `product_hunt`, `skill`, `blog`, `paper`, `event`, `all`
- `--limit N` (optional, default 20): max items to return
- `--query KEYWORD` (optional): case-insensitive substring filter over title + description
- `--json` (optional): emit JSON instead of markdown
## Default interactive flow (no args, or vague request)
This skill covers a lot of ground — 11 feeds spanning news, social, papers, events, and more. If the user invokes it without specifying a feed (e.g., "show me what's new on Agentic Brew", "give me today's AI digest"), do NOT silently default to one feed. Instead, before fetching anything:
1. **Ask the user which categories they want.** Use the host agent's question UI (in Claude Code: `AskUserQuestion` with `multiSelect: true`) so the user can pick any subset of:
`news`, `twitter`, `github`, `reddit`, `youtube`, `product_hunt`, `skill`, `blog`, `paper`, `event`
Plus an `all` shortcut. Show a one-line description of each so the user knows what they're picking. If the user says "everything" or "all", treat as `all`.
2. **Ask the delivery frequency.** Single-select:
- `once` — fetch immediately and return the result.
- `daily` — fetch now AND propose setting up a recurring task. In Claude Code, suggest the `/schedule` skill (cron) or `/loop` (interval). For other host agents, surface their equivalent or tell the user how to re-invoke.
- `weekly` — same idea, weekly cadence.
3. **Ask how much detail to include per item.** Single-select:
- `headlines` — title only. Compact list, just "what happened."
- `summary` — title + the AI-generated summary / overview / engagement stats (whichever the feed provides) + the source link. The default Agentic Brew item shape.
- `detailed` — title + full description (no truncation) + source link + any `content:encoded` inner content (e.g., tweet list for twitter, overview bullets for news) + the `<category>` tags.
To apply the choice: fetch with `--json` internally, then format the items per the chosen detail level. Do NOT pass `--limit` so low that you lose information the user asked for — only `--limit` controls *how many* items, not how *deep* each one goes.
4. Once the user has answered all three, fetch the selected feeds in parallel and present a single combined report grouped by category, formatted at the chosen detail level. If they chose `daily`/`weekly`, ALSO offer to set up the recurring schedule before exiting — don't silently leave it as a one-shot.
If the user provides explicit args (e.g., `/ai-news-radar news --limit 5`), skip the questions entirely and execute directly per the Usage section.
## Steps (direct invocation)
1. Resolve the feed URL from the chosen feed name. If the argument is invalid, abort and tell the user the valid options.
2. Run the fetch + parse one-liner below. It uses the Python stdlib only (`urllib`, `xml.etree`) — no extra installs.
3. Print the result. Default output is a markdown list (`- [title](link) — pubDate · description`). With `--json`, print a JSON array of `{title, link, description, pub_date, categories}`.
## Fetch + parse one-liner
Substitute `FEED`, `LIMIT`, `QUERY`, and `FORMAT` (`md` or `json`) before running.
```bash
FEED="news" # news | twitter | github | reddit | youtube | product_hunt | skill | blog | paper | event | all
LIMIT=20
QUERY="" # empty = no filter
FORMAT="md" # md | json
python3 - "$FEED" "$LIMIT" "$QUERY" "$FORMAT" <<'PY'
import json, sys, urllib.request, xml.etree.ElementTree as ET
FEED, LIMIT, QUERY, FORMAT = sys.argv[1], int(sys.argv[2]), sys.argv[3].lower(), sys.argv[4]
URL = f"https://www.agenticbrew.ai/feed/{FEED}.xml"
req = urllib.request.Request(URL, headers={"User-Agent": "ai-news-radar-skill/1.0"})
with urllib.request.urlopen(req, timeout=30) as r:
xml_bytes = r.read()
root = ET.fromstring(xml_bytes)
items = []
for it in root.iter("item"):
def text(tag):
el = it.find(tag)
return (el.text or "").strip() if el is not None and el.text else ""
title = text("title")
link = text("link")
desc = text("description")
pub = text("pubDate")
cats = [c.text.strip() for c in it.findall("category") if c.text]
hay = f"{title}\n{desc}".lower()
if QUERY and QUERY not in hay:
continue
items.append({
"title": title, "link": link, "description": desc,
"pub_date": pub, "categories": cats,
})
if len(items) >= LIMIT:
break
if FORMAT == "json":
print(json.dumps(items, ensure_ascii=False, indent=2))
else:
for i in items:
d = i["description"]
if len(d) > 200: d = d[:199] + "…"
print(f"- [{i['title']}]({i['link']}) — {i['pub_date']}\n {d}")
if not items:
print("(no items matched)")
PY
```
## Notes
- The feeds always reflect the latest published run_log, so calling this skill twice in the same day usually returns the same items. There is no incremental cursor — caller is responsible for dedup if needed.
- On non-2xx HTTP responses, surface the status code and URL in RED so the user can see which endpoint failed.
- If `xml.etree` cannot parse the body, log a YELLOW warning with the first 200 chars of the response so the user can diagnose (likely a CDN error page, not XML).
- This skill is read-only against a public endpoint — no credentials, no rate limiting on the caller side. Be polite: do not call in a tight loop.