:bug: Continuous cron schedules don't wake hibernated pod, session not reachable in UI

[jezekra1]

Summary

Continuous-mode cron schedules have two related problems reported in Slack:

1. Sessions run in the background but are not reachable from the UI. The schedule status shows success and (at least the first) Slack reply goes out, but there is no visible conversation in the UI for the scheduled runs.
2. Hibernated instances are not reliably woken for continuous schedules. If idleTimeout has put the instance to sleep, subsequent cron fires don't bring the pod back — the schedule ticks but the agent never actually runs. Workaround: set idleTimeout: 0 to keep the pod up permanently.

Repro

• Create an agent instance with default idleTimeout.
• Attach a cron schedule with sessionMode: continuous (e.g. every few minutes).
• Let the pod go idle and be hibernated.
• Wait for the next cron fire.

Expected

• The scheduler wakes the hibernated instance, delivers the trigger, and the resumed session is visible in the UI (linked to the same continuous session on each fire).
• Slack/reply notifications behave consistently across fires.

Actual

• Schedule status is written as success but no conversation appears in the UI for the triggered runs.
• If the pod was hibernated, it is not woken; subsequent fires silently do nothing useful until the user manually opens a side chat (which wakes the pod).
• Only the very first Slack reply was observed — unclear whether the model legitimately chose to stay silent after that, or whether later fires never reached the agent at all.

Suspected area

• packages/controller/pkg/scheduler/scheduler.go — wakeIfHibernated / waitForPodReady before kubectl exec delivery of the trigger file.
• packages/controller/pkg/reconciler/idlechecker.go — interaction between idle hibernation and cron-initiated wake.
• packages/api-server/src/apps/harness-api-server/app.ts handleTrigger — how continuous sessions are persisted / made visible (sessions.findByScheduleId + onSessionCreated).
• packages/agent-runtime/src/trigger-watcher.ts — trigger file processing inside the pod.

Reported by

Radek / Tomas in Slack (2026-04-19). Tomas: "schedules jsou bugged, fungujou na pozadí, ale k té session se nedostaneš." Radek: "musí ten agent běžet, jinak ho to neprobudí — v continuous minimálně, musel jsem vypnout idleTimeout úplně."