← All updates

33-agent system audit

First major architectural audit. 33 agents in parallel asked "what's working, what's missing, what would a serious operator notice." The direction was right; the gaps were in operations and reliability, not in the architecture.

BackendDesktopAndroid auditarchitecture

The first proper architectural audit on Crumb. I asked thirty-three different questions at once, mostly variations on "would a serious operator be happy here" and "what's missing compared to how the gold-standard tools work."

What came back

The architecture and the UI direction were right. The deltas were in the unglamorous parts:

  • Ops: no health dashboard, no resource alerting, no clear "is this thing

recording right now" signal at a glance.

  • Reliability: no chaos-test suite, no automated DB backup, gaps in the

recorder restart story (which became the footage reliability audit a few days later).

  • Polish: lots of small UX edges where the desktop client was technically

correct but felt rough compared to what a long-time operator expects.

What it produced

A backlog. Some of it has since shipped (recorder health dashboard, the size-accounting drift fix, the storage policy redesign, smoother playback-scrub behavior). Most of the rest is on the open roadmap.

Why audits keep happening

Because side projects rot quietly. Running one means deliberately stepping back and asking, with the harshest reading: what's wrong here? This audit set the pattern for everything since.

Full doc: docs/AUDIT-2026-06-15.md.

Don't miss the next one.

Email when something ships. Or grab the RSS.