Org knowledge recall

Institutional memory in a small operator lives in people's heads and scattered chat threads. When the org chart is one person plus agents, that recall problem gets sharper, not softer. I want the history searchable and answerable, so decisions do not get re-litigated.

How it works

The design is the same Slack app already being stood up for the HR alerts, extended. A Socket Mode listener (an outbound WebSocket, so the NAT'd mini needs no public URL) logs channel messages into a local SQLite store with full-text search, and an @-mention answerer bridges to Claude, reusing the same headless-Claude bridge the housekeeping co-pilot already runs on.

• The bot only reads channels it is invited to, a natural privacy boundary
• No DM reading and no audio, ever, since voice is off the table under California recording law
• Gated behind a plain-language staff AI-assistant notice

It is deliberately deferred. A leverage check flagged it as infrastructure for its own sake until there is a concrete recall use case worth the privacy surface. So it is designed, not built.

This is the bet disciplining itself. It would be easy to build because the plumbing already exists, which is exactly why it is on the bench. The test is not "can an agent do this seat's work," it is "does doing it actually buy me leverage." When the answer is no, the right move is to not build it yet.