After my review job succeeds, run a second job to inspect its findings. Keep the follow-up manually runnable.Discover source job ID with
list_jobs, and choose a target with list_execution_targets. Create the destination with required job fields and:
Inspect source job {{trigger-2.sourceJobId}} and run {{trigger-2.sourceRunId}}. Summarize its findings. Use the actual slug. Context includes identifiers, state and completedAt, not the source output. The execution agent must have its own authorized way to retrieve source results (for example its own Factorize MCP access); the conversational client’s authorization is not forwarded. If unavailable, have the conversational agent inspect the source output in the browser or via REST trace pages and manually invoke the destination with output in data instead of claiming automatic output transfer.
Expected result: a successful source run produces a destination run. Invoke the source with a stable key, wait for success, then inspect destination runs and full context to verify source job/run IDs, then inspect output in the browser. Avoid cycles and make prompts safe for repeated events; automatic signals may coalesce.
Available states are succeeded, failed, stopped, and edited. Edited fires on successful saves (including enable/disable), even unchanged saves, not creation or run-state changes. Edited context has source_job_id, event: "edited", and edited_at; it has no source run ID. Only same-tenant enabled subscribers match, and a job cannot subscribe to its own edits. Pending edit events wait while disabled/unmatched; deleting the source or removing the trigger discards them. Verify an edited recipe by saving the source and inspecting the destination context.