Observed arrival · 2026-08-26
RunVouch, the watchdog for unattended AI agents
A monitoring and verification tool for scheduled AI-agent runs, checking whether jobs actually produced their promised results.
Field notes
The documented workflow registers an agent with a cadence, cost cap, and evidence requirement, then wraps an existing cron, systemd, GitHub Actions, or Claude Code command. RunVouch’s examples distinguish a green process exit from a changed output file and identify a retry storm through 41 repeated calls in four minutes. Completed runs are described as JSON records containing status, cost, tokens, tool calls, and evidence verdicts, paired with a SHA-256 leaf.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue