Workflow Foreman
The automation operator that builds, runs, and supervises long-running workflows on your behalf
Workflow Foreman is the long-running automation surface of Hannah Imagineer. It builds n8n-style workflows from intent, runs them on a managed n8n executor, monitors every execution, and escalates exceptions back to Composer for human review.
Overview
Foreman owns the "set it and forget it" half of the platform: scheduled ingestion, multi-step API choreography, cross-system sync, and any automation that needs to survive across days and retries. Workflow definitions are versioned; every execution is persisted in workflow_executions for audit and replay.
Primitives
Intent-to-workflow generation via n8n DSL
Managed n8n executor with persistent run history
Scheduled triggers (cron) and event triggers (webhook)
Exception escalation back to Composer for human review
Versioned workflow definitions with rollback
Native integrations with the rest of the HAIF agent fleet
Outcomes
- 01Replace a backlog of manual cross-system sync jobs
- 02Long-running automations supervised by an agent, not a human
- 03Every workflow run is observable and replayable
Integrations
Autonomy & guardrails
- Workflow edits that change destructive-node permissions
- Scheduled triggers outside the configured cadence window
- Cross-tenant workflow execution
Guardrails & requirements
Guardrails
- Workflow edits go through Composer approval gates
- Every run is persisted with full input/output capture
- Destructive nodes (delete, refund, terminate) require explicit allow-listing per workflow
Requirements
- Managed n8n instance OR self-hosted n8n endpoint
- Outbound HTTPS to the systems the workflow integrates with
Technical specifications
Runtime
- Harness
- Managed n8n executor
- Deployment
- Managed (Hannah Imagineer cloud) · Self-hosted n8n endpoint supported
- Data residency
- Customer-controlled — workflow defs and run history live in the customer Supabase
- License
- Commercial · Hannah Imagineer EULA · n8n Sustainable Use License (managed)
- Version
- v1.0
Models & tooling
- Models
- claude-sonnet-4-6gpt-5
- Tooling
- n8n-dsl-compilerexception-routerrun-history-store
Reliability & limits
- Latency SLA
- ~3s p50 workflow compile · execution time bounded by integrated systems
- Rate limits
- 500 executions/min per workspace
Security & compliance
- Auth model
- Composer SSO · per-workflow credentials in HashiCorp Vault or n8n credential store
- Compliance
- SOC 2 Type II (in process)GDPR
Evals & benchmarks
Public workflow-generation benchmarks are still maturing as a category. The internal eval scores compile success, first-execution success, and time-to-stable on a held-out workflow-intent corpus; we''ll publish numbers once the eval methodology document lands alongside the Composer-native benchmark release.
Architecture notes
Memory architecture
Workflow definitions and run history are persisted in the customer Supabase (workflow_executions table). Foreman reads cross-agent context from Shared Mind when constructing or repairing a workflow.
Context strategy
Context for workflow generation is assembled from the customer''s existing workflow library (via Shared Mind) plus the n8n node catalog. Per-execution context is just the trigger payload — Foreman keeps the agent step minimal once the workflow is compiled.
Evaluation
Evaluation runs against a held-out workflow-intent corpus scored on compile success, first-execution success, and time-to-stable (how many human-supervised fixes before the workflow runs cleanly).