heartbeat

Name: heartbeat
Author: MagicCube

489stars57forks

How to install heartbeat?

magiccube/agentara/heartbeat

$npx -y skills add MagicCube/agentara --skill heartbeat

Installs into the current project.

Summary

Proactive intelligence pulse — scans user's memory files (USER.md, SOUL.md) to identify current interests, investments, projects, and concerns, then performs targeted web searches to find noteworthy updates. Returns a concise briefing if something deserves attention, or `[SKIPPED

Files · 1

View on GitHub

SKILL.md

1	# Heartbeat — Proactive Intelligence Pulse
2
3	You are performing a heartbeat check: scanning the user's world for things that matter to them right now, and deciding whether any of it is worth their attention.
4
5	## Philosophy
6
7	The user's time and attention are sacred. A heartbeat that cries wolf is worse than no heartbeat at all. Only surface information that is actionable, time-sensitive, or significantly changes the user's understanding of something they care about. When in doubt, return `[SKIPPED]`.
8
9	## Step 1: Read Memory
10
11	Read the user's memory files to build a profile of current interests and concerns:
12
13	Extract a prioritized list of watchlist topics. Typical categories:
14
15	\| Category \| Examples \| Alert Threshold \|
16	\|----------\|---------\|-----------------\|
17	\| Investments \| Stock positions, crypto holdings \| Price moves >5% in a day, major company news (earnings, lawsuits, leadership changes), analyst upgrades/downgrades \|
18	\| Current project \| Active codebase, architecture decisions \| Major version releases of key dependencies, breaking changes, security vulnerabilities \|
19	\| Career/Industry \| Employer news, role-relevant trends \| Company announcements, industry shifts directly affecting the user's role \|
20	\| Tech stack \| Frameworks, tools, languages used daily \| New major releases, deprecation notices, critical CVEs \|
21	\| Upcoming events \| Travel plans, deadlines \| Weather alerts, schedule changes, booking reminders within 7 days \|
22	\| Personal interests \| Photography gear, anime, content creation \| Notable releases or events only (not routine content) \|
23
24	## Step 2: Read News
25
26	Fetch the following sources directly by URL — no search needed, just fetch:
27
28	\| Source \| URL \| Notes \|
29	\|--------\|-----\|-------\|
30	\| Hacker News \| https://hnrss.org/newest \| RSS feed \|
31	\| Product Hunt \| https://www.producthunt.com/feed \| RSS feed — high weight: today's top launches often contain directly actionable tools or competitors \|
32	\| GitHub Trending \| https://github.com/trending?since=daily \| Daily trending — high weight: surfacing repos that match the user's stack or interests is high-value signal \|
33	\| Google News (CN) \| https://news.google.com/rss?hl=zh-CN&gl=CN&ceid=CN:zh-Hans \| RSS feed \|
34
35	Weighting guidance: Product Hunt and GitHub Trending are real-time ranked signals — their top items reflect what's genuinely popular today. Prioritize items from these two sources over generic news articles when relevance is comparable. A trending repo or a #1 Product Hunt launch that matches the user's stack or project is almost always worth mentioning.
36
37	## Step 3: Search
38
39	For each high-priority watchlist topic (aim for 3-5 topics, no more than 8), perform a focused web search. Use queries that are specific enough to surface real news, not evergreen content.
40
41	Good search patterns:
42	- `"ByteDance" news March 3 2026` — catches employer news
43	- `小荷 AI health news March 3 2026` — catches project news
44	- `"BABA stock" March 3 2026` — catches daily movers
45	- `"LangChain" release 2026` — catches recent releases
46	- `"Claude" OR "Anthropic" announcement` — catches AI tooling updates
47
48	Skip topics where nothing meaningful could have changed since the last check (e.g., stable personal preferences, unchanging habits).
49
50	## Step 4: Evaluate — The "Worth Interrupting" Test
51
52	### Hard filter: Date verification (run this FIRST, before anything else)
53
54	For every item found in Step 2 or Step 3, you must verify its publication date by checking the article page or search result metadata. Record the date explicitly.
55
56	- If the publication date is within the last 48 hours (T or T-1 relative to today): eligible, proceed to the Worth Interrupting test below.
57	- If the publication date is older than 48 hours, or cannot be determined: discard immediately, do not include.
58
59	This is a hard gate — relevance does not override staleness. A perfectly relevant article from last week is still discarded. When in doubt about a date, discard.
60
61	### Worth Interrupting test (only for items that passed the date filter)
62
63	For each remaining finding, apply this filter:
64
65	1. Is it genuinely new? (not something the user already knows from memory or the previous briefing)
66	2. Is it actionable or time-sensitive? (the user should do somethin

People who liked this also viewed

6 picks

Type

heartbeat

489stars57forks

View on GitHub

How to install heartbeat?

magiccube/agentara/heartbeat

$npx -y skills add MagicCube/agentara --skill heartbeat

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Heartbeat — Proactive Intelligence Pulse
2
3	You are performing a heartbeat check: scanning the user's world for things that matter to them right now, and deciding whether any of it is worth their attention.
4
5	## Philosophy
6
7	The user's time and attention are sacred. A heartbeat that cries wolf is worse than no heartbeat at all. Only surface information that is actionable, time-sensitive, or significantly changes the user's understanding of something they care about. When in doubt, return `[SKIPPED]`.
8
9	## Step 1: Read Memory
10
11	Read the user's memory files to build a profile of current interests and concerns:
12
13	Extract a prioritized list of watchlist topics. Typical categories:
14
15	\| Category \| Examples \| Alert Threshold \|
16	\|----------\|---------\|-----------------\|
17	\| Investments \| Stock positions, crypto holdings \| Price moves >5% in a day, major company news (earnings, lawsuits, leadership changes), analyst upgrades/downgrades \|
18	\| Current project \| Active codebase, architecture decisions \| Major version releases of key dependencies, breaking changes, security vulnerabilities \|
19	\| Career/Industry \| Employer news, role-relevant trends \| Company announcements, industry shifts directly affecting the user's role \|
20	\| Tech stack \| Frameworks, tools, languages used daily \| New major releases, deprecation notices, critical CVEs \|
21	\| Upcoming events \| Travel plans, deadlines \| Weather alerts, schedule changes, booking reminders within 7 days \|
22	\| Personal interests \| Photography gear, anime, content creation \| Notable releases or events only (not routine content) \|
23
24	## Step 2: Read News
25
26	Fetch the following sources directly by URL — no search needed, just fetch:
27
28	\| Source \| URL \| Notes \|
29	\|--------\|-----\|-------\|
30	\| Hacker News \| https://hnrss.org/newest \| RSS feed \|
31	\| Product Hunt \| https://www.producthunt.com/feed \| RSS feed — high weight: today's top launches often contain directly actionable tools or competitors \|
32	\| GitHub Trending \| https://github.com/trending?since=daily \| Daily trending — high weight: surfacing repos that match the user's stack or interests is high-value signal \|
33	\| Google News (CN) \| https://news.google.com/rss?hl=zh-CN&gl=CN&ceid=CN:zh-Hans \| RSS feed \|
34
35	Weighting guidance: Product Hunt and GitHub Trending are real-time ranked signals — their top items reflect what's genuinely popular today. Prioritize items from these two sources over generic news articles when relevance is comparable. A trending repo or a #1 Product Hunt launch that matches the user's stack or project is almost always worth mentioning.
36
37	## Step 3: Search
38
39	For each high-priority watchlist topic (aim for 3-5 topics, no more than 8), perform a focused web search. Use queries that are specific enough to surface real news, not evergreen content.
40
41	Good search patterns:
42	- `"ByteDance" news March 3 2026` — catches employer news
43	- `小荷 AI health news March 3 2026` — catches project news
44	- `"BABA stock" March 3 2026` — catches daily movers
45	- `"LangChain" release 2026` — catches recent releases
46	- `"Claude" OR "Anthropic" announcement` — catches AI tooling updates
47
48	Skip topics where nothing meaningful could have changed since the last check (e.g., stable personal preferences, unchanging habits).
49
50	## Step 4: Evaluate — The "Worth Interrupting" Test
51
52	### Hard filter: Date verification (run this FIRST, before anything else)
53
54	For every item found in Step 2 or Step 3, you must verify its publication date by checking the article page or search result metadata. Record the date explicitly.
55
56	- If the publication date is within the last 48 hours (T or T-1 relative to today): eligible, proceed to the Worth Interrupting test below.
57	- If the publication date is older than 48 hours, or cannot be determined: discard immediately, do not include.
58
59	This is a hard gate — relevance does not override staleness. A perfectly relevant article from last week is still discarded. When in doubt about a date, discard.
60
61	### Worth Interrupting test (only for items that passed the date filter)
62
63	For each remaining finding, apply this filter:
64
65	1. Is it genuinely new? (not something the user already knows from memory or the previous briefing)
66	2. Is it actionable or time-sensitive? (the user should do somethin

People who liked this also viewed

6 picks

Type