Category 05 of 12
The Workflow Engine
Introducing the Workflow Engine — the orchestration substrate behind Agentic Apps: the recovery ladder, the SWIFT verdict that separates completed from accomplished, and the node reference.
Pages in this category
Most automation reports success the moment every step fires. The Workflow Engine refuses that shortcut — it checks the world, heals what it can, and names honestly what it can't.
The exact loop: draining the ready queue up to a parallelism cap, a three-way dispatch decision per node, and the counter that stops a run from settling mid-dispatch.
The five-stage discipline behind every hardened workflow, from declaring how success will be verified to a verdict engine that never trusts a self-report.
Four real mechanisms working together — a recovery ladder, a law against silent restructuring, a pattern-recognizer for repeat failures, and a safe way to change a live run — not a single feature.
The SWIFT verdict engine probes the real world after a run settles instead of trusting a self-report, distinguishing accomplished from merely completed.
How a failed node recovers: deterministic rebind, capability-aware remedy, config-aware surfacing, certified structural repair, then honest escalation.
A proven, blessed workflow graph is never autonomously restructured — repair requires certification and a rollback path.
Inside WorkflowEngine: the ReadyQueue, the WaitingInputBuffer, and what happens on every tick from dispatch to snapshot.
How a run-scoped Objective declares the acceptance checks that define done, verified against real outputs rather than a self-report.
Converge iterates a cohort sub-graph until a continuation policy stops it; Pursue is converge with ASSESS and REFLECT turned on.
A graded 0..1 progress signal for stall detection, and a bounded self-critique pivot ladder for when a loop stalls.
The doctrine injected into both the synthesis and reviewer prompts, teaching agents to design for failure and control flow, not just structure.
Static named failure/control-flow patterns, plus a living, workspace-scoped playbook of failure-mode-to-fix lessons recalled at build time.
The deterministic enforcement layer that audits an authored graph for robustness before it ever runs.
Detects repeated workflow failure patterns, writes the lesson into the Brain, and emits an operator-visible proposal.
Mid-run graph changes must pass the same contracts as authoring — no data-coupling break, no gutted capability, never a regression below the graph's proven green.
The regression gate that runs cold-start build tasks against a seeded workspace and grades the actual outcome — pass@k versus pass^k.
Resolves a single capability URN onto one execution path, so chat, workflow, and MCP all delegate through the same trusted handlers.
Manual, cron (natural-language to UTC), webhook, and persistent-listener triggers — how a run actually starts.
Restarting a run from a node, a failed branch, an edited node, or a checkpoint — deterministic partial replay.
How parallel agent work gets its own git worktree, so a swarm or a converge cohort never steps on another branch's changes.
router, merge, parallel, wait, checkpoint, phase_gate, subflow, loop, stop_error, return_output — every control-flow node explained.
agent_task, agent_session, agent_swarm, converge, pursue, planner, evaluator, guardrails — every intelligence node explained.
transform, code, data_query, http_request, aggregate_window, knowledge, browser, artifact_save, mcp, human_input — every data/compute/IO node explained.