Agent workflows

Design repeatable media-processing jobs

Use stable job identities and explicit completion records so retries produce dependable derivatives rather than duplicate work.

Make a repeated request lead to a clear result

Media jobs can be interrupted, retried or triggered twice. A dependable workflow knows whether the requested output already exists, which input produced it and whether it passed validation. Suite gives authorized tools access to shared files; the job system must manage its own execution and completion records.

Define a job identity from the source revision, operation and relevant configuration. Two requests to produce the same thumbnail set from the same source with the same settings should be recognizable as the same logical work. A changed source or model should produce a different identity.

Separate work in progress from completed output

Have the worker write into an assigned working area or a uniquely named candidate destination. Mark completion in the job system only after the required output checks pass. Do not assume a visible filename proves that a writer has finished.

If a finalization step uses rename or another filesystem operation, verify its behavior on the actual access path with Suite. Do not design around an unverified cross-interface atomicity guarantee. A manifest that records validated output locations can be a practical application-level boundary.

Treat triggers as requests to inspect state

AWS documents S3 event notifications as at-least-once and potentially out of order. An event-driven object workflow should therefore check its job record rather than assume every notification represents new work. The same design discipline helps when a human or scheduler repeats a request.

Record queued, running, succeeded and failed states with enough evidence to explain them. Decide how a stale running job becomes eligible for recovery, and make sure two workers cannot both claim ownership through an untested file-locking assumption.

Validate one interruption

Use disposable test media and stop a job at an agreed point. Restart it and confirm that it either reuses a validated result or creates one clearly identified replacement. Check that no partial file is presented as approved and no source original is overwritten.

Then change the processing configuration and verify that the new output is distinguishable. This exercise tests the workflow rather than merely a successful first run. Suite supports the shared input/output path, while the processing system supplies repeatability. Explore the agent guide and keep the first job small enough to inspect end to end.

Put the idea to work

Start with your workflow.

Bring shared files into your familiar applications with Suite Managed, or connect your own supported storage with S3 Native. S3 Native requires at least 20 TB of active storage.