[JoeZhang-0x000]

Closes #790

Summary

• Add optional on-demand monitoring for direct SSH runs, allowing agents to control status refreshes and log collection.
• Preserve automatic monitoring defaults and configurable intervals.
• Persist monitoring policy and cancellation intent; handle restart recovery and final log collection.
• Add saved defaults and explicit UI refresh controls. Keep SSH keepalives while removing periodic helper heartbeats and redundant log writes.

Test plan

• Rust tests: 1,331 passed, 3 ignored; fmt and Clippy passed.
• Mock SSH regressions cover idle behavior, cancellation races, recovery, final logs, and container scope.
• SSH helper tests: 14 passed.
• UI tests: 438 passed; UI build and style checks passed.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge; no actionable issues were found.

Summary

Adds optional on-demand monitoring for direct SSH runs while keeping automatic monitoring as the default.

• Direct SSH runs can keep automatic checks or use a saved interval.
• Direct SSH runs can wait for an explicit status or log refresh.
• Run details show cached SSH observations and offer refresh controls.

No actionable issues were found. This review checked code and existing tests; it did not run the test suites.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Submit direct SSH run] --> B[Save monitoring policy]
  B --> C{Monitoring mode}
  C -->|Automatic| D[Supervisor checks status and collects logs]
  C -->|On demand| E[Wait for explicit refresh]
  E --> F[Check status once]
  F --> G{Collect logs requested?}
  G -->|Yes| H[Copy up to 1 MiB within 20 seconds]
  D --> I{Terminal and logs complete?}
  H --> I
  I -->|Yes| J[Queue remote cleanup]
  K[Cancel requested] --> L[Save intent]
  L --> M[Make one bounded control attempt]

Loading

_{Reviews (1) · Last reviewed commit: "feat(ssh): add optional on-demand monito..." · Reviewed by Greptile}

[paperclover]

thanks for this, it's a big and careful patch. #862 replaces the per-run orx supervise monitor it builds on: one engine inside orx up now watches every run over a single long-poll connection per host, so the repeated polling this was cutting down is gone. there's no orx refresh or interval setting on the engine yet, so #790 stays open for that, and we'll build it on the engine ourselves rather than asking you to port this.