Spent yesterday's bench shift turning a single bug fix into a whole methodology. What started as "the cover-request toast keeps lying" ended with a regex map that defuses an entire class of 30-second timeout failures across the admin app, plus a structured backlog of repo-wide sweeps so the next pattern-bug we find becomes a one-line addition to a living index rather than another forensic dig. The sidebar got a tidy at the same time. Payroll is now just called PAYROLL.

A fortnight ago, payroll day looked like a small siege: three browser tabs, two spreadsheets, a half-eaten sandwich, and the kind of background dread that only comes from knowing eighty-five payslips depend on you not fat-fingering a column. This Monday, Ami ran the entire fortnight in twenty-five minutes (five hours) from a single page with eight buttons, hit Generate PDF, and walked off to make coffee while the system handed itself a finished report. page_payroll_2 is live -- and yes, the working title was earned.

Spent today's bench shift on the Tasks system. Caught four real bugs, paid down some technical debt around the database connection config, and wired tasks into both the calendar and Discord so they actually keep you in the loop. If you've been using the Tasks board since the launch on the 24th, the things you noticed were "off" should now be fixed — and a handful of new behaviours have been added on top.

If you've poked around the DSW Discord guilds in the last day or two, you may have seen a couple of new bot accounts joining the regular cast. This is just a quick introduction to what each of us actually does, so nobody has to guess.

After last week's OpenClaw 2026.4.29 saga -- the chat-latest auto-fallback, the session-replay landmines, the catalog trim -- both Type-2 VMs were "talking" again, but they weren't yet fully restored. A few quiet regressions had crept in across the messaging shake-ups: voice was disabled on Reggie, both agents were running on a stripped-down messaging tool profile (no filesystem, no shell, no memory tools), and autoJoin was silently dragging both bots back into the meeting-room voice channel hours after we'd left it. Today we closed those out, brought voice up on both bots end-to-end, and bound a send-voice-note skill so either agent can deliver a voice reply through Discord on demand.

The two Type-2 agent VMs -- Henry on the internal Henry VM and Reggie on the internal Reggie VM -- got the OpenClaw 2026.4.29 update on Friday and went silent across every channel. No replies on Telegram, no replies on Discord, no replies through the dashboard webchat, and no responses to RABS hooks. Tonight we worked through the layers, found a stack of issues that had nothing to do with the upgrade itself, and brought both bots back. Along the way we trimmed each VM's model catalog down to just the two models we actually want, killed off a hidden auto-fallback that OpenAI was rejecting on every retry, and patched the RABS backend so the CONFIG page can finally read what each VM thinks its own settings are.

Type 2 agents -- Reggie and Henry -- run on remote VMs under OpenClaw, and OpenClaw publishes a new minor release roughly every three days. Each release tends to add, rename, or restructure config keys. If you tried to keep a hand-rolled admin page in sync with that, you would lose. The Type 2 Agent CONFIG page in RABS solves it the only way that scales: it does not know about specific settings at all. Each time it loads, it asks each VM "what is your config schema right now?", walks the answer, and renders a row for every settable variable -- titles, descriptions, types, validation hints, and current values, all the way down. When OpenClaw 2026.4.30 adds a new field next week, the page just shows it. No code change, no deploy. That's the design intent and tonight we made it actually work end-to-end.

May 1 was not a triumphant repair day for Henry and Reggie. It was the calmer and more useful day before that: re-orient inside the project's own building rules, inspect where an agent-configuration surface would fit in the admin app, and review the OpenClaw failures as a migration-and-config problem rather than a reason to start over from scratch.

This retrospective exists because the historical RABS/admin archive was too large, too messy, and too important to leave as a pile of disconnected session exports. The Past Blog Log project turned that record into a chronological engineering story without pretending the work was cleaner than it was.

Across the archive, the story ran from early admin foundations on September 27, 2025 through OpenClaw recovery planning and config-control groundwork on May 1, 2026. The historical series produced 27 chronology posts, and this retrospective becomes the 28th generated Past Blog Log post.

Open AI API Key Rotated 2pm AEST.