hey hn, i built murmur. its a cron daemon that runs claude code
sessions on a schedule.
you write a HEARTBEAT.md file in your repo with yaml frontmatter
for config and a prompt below:
---
name: competitor watch
cron: 0 9 * * MON
agent: claude-code
model: sonnet
timeout: 10m
---
fetch https://competitor.com/changelog.
compare against ~/tracking/competitor-last.md for new entries.
for each new feature, check our issue tracker and think about
whether it makes sense for our product given our roadmap.
only if it genuinely adds value: open an issue with reasoning.
update competitor-last.md. if nothing new, HEARTBEAT_OK.
murmur reads the file, runs claude on schedule, claude uses its
tools to do the work. the file is version controlled, similar
format to skills files.
you could run claude in github actions too but murmur runs on your
own machine. a mac mini or raspberry pi in the corner, full access
to your local files and network.
got the idea from openClaw's heartbeat but didn't want a full
platform, just the scheduling part.
you write a HEARTBEAT.md file in your repo with yaml frontmatter for config and a prompt below:
murmur reads the file, runs claude on schedule, claude uses its tools to do the work. the file is version controlled, similar format to skills files.
you could run claude in github actions too but murmur runs on your own machine. a mac mini or raspberry pi in the corner, full access to your local files and network.
got the idea from openClaw's heartbeat but didn't want a full platform, just the scheduling part.
bun + effect-ts, single binary, no cloud.
brew install t0dorakis/murmur/murmur
or: npx skills add t0dorakis/murmur --skill heartbeat-cron
i run about 6 of these for competitor analysis, issue triage, weekly paper research, daily briefs. curious what pepole would use it for.