Real-time monitoring dashboard for Proxmox VE, PBS, Docker, Kubernetes, TrueNAS and vSphere. Self-hosted, with alerts for failed backups, full disks, unhealthy containers and offline nodes.
Development and Automation Transparency
I maintain Pulse, and I build it with heavy use of coding agents and automation. This page explains how, so it doesn't need repeating in every commit, release or issue.
Who does what
- I maintain Pulse: I set direction, decide what gets built, and answer for everything that ships. I don't claim to have written every line.
- Coding agents write code, tests and documentation from requirements I set.
- Automation handles routine triage and fixes within limits I set. Its issue
replies and commits come from
pulse-triage[bot]; repository workflows appear asgithub-actions[bot]. Each problem in a mixed report gets its own answer under the triage contract.
I handle private support email myself. Automation does not read a support mailbox or send email. Keep credentials and private diagnostics out of public threads.
How changes land and ship
Every change reaches main through a pull request. Automated changes receive
independent review, and every change must pass the repository's required checks
before merging. Nobody can bypass this, me included.
Releases follow a fixed weekly train: daily betas, then a release candidate that
soaks for 24 hours and is promoted unchanged to stable unless a release-blocker
issue stops it. Stable patches are not yet available, so fixes reach stable in
the next minor release; see limited stable patches.
Full rules: Releases and update channels.
A fix merged to main is not yet a published fix. Replies name a version only
after checking its released source.
Issue resolution and retesting
An issue is resolved when the reporter confirms the fix, or the fix is reproduced with no reported symptom left unresolved. A merged fix or a workaround alone is not enough.
Once the fix ships in a stable release, the issue gets a reply naming that
version and asking for a retest. After 14 days with no reply, and nobody else
reporting the same problem, the issue may be closed with the
closed-needs-retest-timeout label. That is not confirmed resolution, and
it can be reopened. Feature requests, needs-human issues and unexplained
symptoms are never closed this way.
Still broken? Comment on the original issue rather than filing it again. Say which symptom remains and which version you're on; add redacted diagnostics if you have them. Don't repeat an outage, update or backup just to retest.
Authority and responsibility
Routine fixes and issue replies don't need my sign-off each time, and the release train publishes on schedule without per-release approval; no model chooses what goes in a release or when. None of this skips review, soak or release checks. I can switch any of it off, and if something automated is wrong, it's my bug to fix.
How the work should be judged
By the software: is it correct, secure, tested and maintainable? The source, issues, tests and releases are all public.
Older activity may predate this policy. For Pulse's own AI features, such as Patrol and MCP, see AI.md.