Watchdog
Nudges stuck work to its owner, then escalates to the architect.
Platform · Self-Healing Operations
It keeps itself running.
Stuck work gets nudged, outcomes get checked, and the platform can propose fixes to itself.
Live in productionPeace
Watchdogsim time 09:00next outcome check in 15 minillustrative
Task board
Escalation ladder
- 01Detectno progress for 20 min
- 02Nudge the ownerthe role that holds the task
- 03Escalate to architectstill stuck after 40 min
- 04Incident and root causefix task with a regression test
App heartbeatHealthyGuardian restarts the app if it hangs
Outcome checks
What is Self-Healing Operations in agnte?
agnte self-healing operations keep a long-running fleet healthy. A watchdog nudges stuck work to its owner and then the architect, outcome checks run every 15 minutes on throughput, capacity, merges, dev server freshness, testers and disk trend, and a root cause inside the platform can become a fix task with a regression test that you approve.
Capabilities
What Self-Healing Operations does
Peace
Outcome checks every 15 minutes
Throughput, capacity, merges, dev server freshness, testers, disk trend and request floods.
Self-improvement
Platform root causes become fix tasks with regression tests; you approve the merge. Off by default.
Guardian
External watchdog restarts the app if it hangs.
Proof from production
Running today, measured.
14 dayscontinuous operation
903/903tasks done in a 1-hour soak
Source: Production run 2026-09-28 to 2026-10-12; soak test.
How it works
How Self-Healing Operations works
- Check outcomes
- Detect drift
- Nudge or escalate
- Propose a fix
FAQ
Questions about Self-Healing Operations
Can agnte change its own code?
Only with your approval. Self-improvement is off by default, and every fix needs your merge.
How often are outcomes checked?
Every 15 minutes.
What happens to stuck work?
The watchdog nudges the owner, then escalates to the architect role.
Related