Recurring-run scheduling (roadmap E11). Translates user-friendly presets,
intervals, one-shots, and raw cron expressions into persisted next_run_at
instants for the background scheduler.
Outcome of a coding-agent turn run via FlowApp::run_agent_turn: the run
summary plus the filesystem edits the agent proposed (staged, not yet
applied) for the IDE to review and apply.
Result of an autonomous agentic loop (FlowApp::agentic_run_loop).
Serialised camelCase for the frontend, which loads finalDsl onto the
canvas and renders the per-iteration steps as a run log.
Outcome of a hybrid flow-graph generation. Carries the DSL text plus a
machine-readable record of which path produced it so the UI can
render a “local” / “cloud (fallback)” tag without re-deriving the
information.
If node performs a destructive operation, return a short human-readable
description of it; otherwise None. Used by the per-step confirmation gate
(roadmap E1). Detection is intentionally conservative — it flags the clear
data-loss cases (deleting files, rm, git push/reset --hard/clean,
kubectl delete, drop table/database, truncate, mkfs, dd if=,
shutdown/reboot) rather than guessing at every possible command.
A pre-apply advisory about a node in a proposed flow, surfaced in the review
Host machine snapshot for the Model Hub device-compatibility check. RAM/disk
are in binary GB (GiB) so they compare directly against the catalog’s
minRamGb / minDiskGb.
Tagged enum recording which side of the hybrid Auto path produced the
DSL. The internally-tagged JSON form ({ "kind": "local"|"cloud", ... })
is what the frontend’s ModelChoice discriminator expects to receive.
Where a flow’s shell / cli / fs nodes run, before any explicit
per-node cwd / workspaceRoot. The default is edition-specific so the
shared core serves desktop, server, and CLI from one resolution path.
Safety gate for the autonomous agentic loop: the hard ceiling on
generate→run→observe→re-plan cycles. The loop stops the moment a run has
zero failures; this bounds the unhappy path so the agent can’t spin
indefinitely (and can’t rack up unbounded model/compute cost).