Rodney

Overview

Rodney is an autonomous OpenClaw agent I built to keep Thirteen honest. Every night it pulls each customer's last 24 hours of WhatsApp messages and compares what people asked for against what actually got recorded: timesheets, invoices, change orders. Mismatches, missing entries, phantom data, and agent miscommunication get ranked by severity and posted to Discord as a per-organization report before anyone on the team is awake.

How it works

• Eleven read-only tools over Thirteen's Convex backend, registered as an OpenClaw plugin. Rodney can discover organizations, pull daily summaries, and drill into a single user's conversations or timesheet entries, but it can never write.
• Intent versus state. The core check is simple to state and hard to do well: did the system do what the person meant? Rodney reads the conversation, reads the records, and flags the gap.
• Pull request review. Once the nightly audit was stable, I extended Rodney to review the team's pull requests with the same backend context.

Why it mattered

With three contractors live and a two-person team, we could not manually check every WhatsApp thread. Rodney turned "I think the agent logged that wrong" into a report with a link, every morning.