Repository navigation
failproofai 1.0.15
This release stops OpenClaw tool calls from being blocked as "failproofaid could not be reached" while the daemon is running normally.
Fixes
OpenClaw
- Tool calls are no longer blocked while the gateway is busy. The plugin timed its wait for the daemon on the gateway's own event loop, which every agent shares. When the gateway was busy, for example while scheduled agents started together, a reply that had already arrived was read too late, and the call was blocked. A busy gateway no longer counts as a failure. (#1024)
- A failed connection is retried once. If the plugin can't reach the daemon, it tries again after 100 ms before blocking the call, which rides out a daemon restart. A call is never evaluated twice.
- Blocks say why in the gateway log. Each blocked call logs its cause, such as
ENOENTorECONNREFUSED, with the elapsed time and the number of attempts. The message the agent sees is unchanged.
Dependencies
posthog-node5.73.0 andsql.js1.16.0, and Next.js 16.4.6 in the bundled dashboard.picomatchhas no fixed release yet for GHSA-fc5q-39xc-vjpf, so it is allow-listed until 2026-12-18. It ships only inside the dashboard's Next.js bundle, where no request can supply the pattern the advisory needs.
Upgrading
- Run
npm i -g failproofai@latest. - Run
failproofai update. - If you use OpenClaw, restart the gateway (
openclaw gateway restart) so it loads the new plugin.
Good to know
- A daemon that is down still blocks, quickly. A missing daemon is detected at once, and the call is blocked after one retry.
- A daemon that never answers blocks after 30 seconds. That limit covers the whole request, including the retry.