Workflow Orchestrator

Workflow Orchestrator — AEM as a Cloud Service

Purpose

This is the master entry point for all AEM Workflow tasks on Cloud Service — spanning both development (building workflows) and production support (debugging and triaging workflow issues). Read this skill first. It classifies the user's request and routes to the right sub-skill.

How to Use This Skill

1. Read the user's request carefully
2. Classify it using the Task Classifier table below
3. Load the identified sub-skill's SKILL.md and its references
4. For development tasks, always load the workflow-foundation references alongside the sub-skill references
5. For production-support tasks, the debugging and triaging skills are self-contained

---

Task Classifier

Development Skills

| User Says / Asks | Sub-Skill to Load | |---|---| | "Create a workflow model", "Add steps to a workflow", "Design an OR split", "I need a parallel review" | workflow-model-design | | "Implement a custom process step", "Write a WorkflowProcess", "Create a ParticipantStepChooser", "Dynamic participant" | workflow-development | | "Start a workflow from code", "Trigger a workflow via API", "Use Manage Publication with a workflow", "HTTP REST API to start a workflow" | workflow-triggering | | "Configure a launcher", "Auto-start on asset upload", "Launcher not firing", "cq:WorkflowLauncher", "Overlay an OOTB launcher" | workflow-launchers | | "How do workflows work?", "Explain workflow architecture" | Load workflow-foundation references only |

Production Support Skills

| User Says / Asks | Sub-Skill to Load | |---|---| | "Workflow is stuck", "Why isn't my workflow advancing?", "No work item", "Workflow failed", "Step shows error" | workflow-debugging | | "Task not in Inbox", "User can't see work item", "Permissions error on workflow" | workflow-debugging | | "Thread pool exhausted", "Auto-advancement not working", "Queue backlog", "Sling Jobs stuck" | workflow-debugging | | "Repository bloat", "Too many workflow instances", "Purge not working", "Stale workflows" | workflow-debugging | | "What workflow errors on host X?", "Workflow activity for the past N hours", "What should I collect?" | workflow-triaging | | "Classify this workflow ticket", "What Splunk query should I use?", "What logs do I need?" | workflow-triaging | | "Why did workflow X fail? Show me the error.", "Failure details for model Y" | workflow-triaging |

Routing heuristic:

• Building/implementing workflows → development skills (workflow-model-design, workflow-development, workflow-triggering, workflow-launchers)
• Deep troubleshooting (decision trees, config checks, thread analysis, remediation) → workflow-debugging
• Incident classification (symptom → runbook, log patterns, Splunk, data gathering) → workflow-triaging
• When both debugging and triaging apply, start with workflow-triaging to classify, then workflow-debugging for resolution

---

Reference Loading Order

For every workflow task on Cloud Service, load in this order:

Step 1: Always load these foundation references

workflow-orchestrator/references/workflow-foundation/architecture-overview.md
workflow-orchestrator/references/workflow-foundation/api-reference.md
workflow-orchestrator/references/workflow-foundation/jcr-paths-reference.md
workflow-orchestrator/references/workflow-foundation/cloud-service-guardrails.md
workflow-orchestrator/references/workflow-foundation/quick-start-guide.md

Step 2: Load the sub-skill's SKILL.md

workflow-model-design/SKILL.md         ← for model design tasks
workflow-development/SKILL.md          ← for Java implementation tasks
workflow-triggering/SKILL.md           ← for start/trigger tasks
workflow-launchers/SKILL.md            ← for launcher config tasks
workflow-debugging/SKILL.md            ← for production debugging tasks
workflow-triaging/SKILL.md             ← for incident triage tasks

Step 3: Load the sub-skill's topic references

workflow-model-design:

workflow-model-design/references/workflow-model-design/step-types-catalog.md
workflow-model-design/references/workflow-model-design/model-xml-reference.md
workflow-model-design/references/workflow-model-design/model-design-patterns.md

workflow-development:

workflow-development/references/workflow-development/process-step-patterns.md
workflow-development/references/workflow-development/participant-step-patterns.md
workflow-development/references/workflow-development/variables-and-metadata.md

workflow-triggering:

workflow-triggering/references/workflow-triggering/triggering-mechanisms.md
workflow-triggering/references/workflow-triggering/programmatic-api.md

workflow-launchers:

workflow-launchers/references/workflow-launchers/launcher-config-reference.md
workflow-launchers/references/workflow-launchers/condition-patterns.md

workflow-debugging:

workflow-debugging/SKILL.md
workflow-debugging/reference.md

workflow-triaging:

workflow-triaging/SKILL.md

---

Cloud Service Production Support Constraints

| Constraint | Detail | |---|---| | No JMX | No retryFailedWorkItems, countStaleWorkflows, restartStaleWorkflows, purgeCompleted via JMX | | Retry failed items | Inbox Retry only | | Stale detection | Custom API/script only | | Purge | Purge Scheduler (OSGi config in Git) | | Log access | Cloud Manager → Environments → Logs (download / streaming) | | Thread dumps | Developer Console or support request | | Config changes | Code in Git + Cloud Manager pipeline deploy |

---

AEM Cloud Service Guardrails Summary

Before doing anything, apply these non-negotiable constraints:

| Rule | Detail | |---|---| | /libs is immutable | Never write to /libs; use /conf/global/ or /apps/ overlays | | Model design-time path | /conf/global/settings/workflow/models/ | | Model runtime path (for API calls) | /var/workflow/models/ | | Launcher config path | /conf/global/settings/workflow/launcher/config/ | | Service users | Always use workflow-process-service sub-service; never admin credentials | | OSGi annotations | Use DS R6 (@Component, @Reference from org.osgi.service.component.annotations) | | Deploy via | Cloud Manager pipeline — no Package Manager in production | | No javax.jcr.Session.loginAdministrative | Use ResourceResolverFactory.getServiceResourceResolver() |

Full detail: references/workflow-foundation/cloud-service-guardrails.md

---

Quick Architecture Recap

Author tier
  │
  ├── Content Author
  │     └── triggers via: Timeline UI / Manage Publication
  │
  ├── Custom code (OSGi service / event handler / scheduler)
  │     └── triggers via: WorkflowSession.startWorkflow()
  │
  └── Workflow Launcher (cq:WorkflowLauncher)
        └── triggers via: JCR observation events
              ↓
        Granite Workflow Engine
              ↓
        Workflow Instance (/var/workflow/instances/)
              ↓
        Steps executed as Sling Jobs:
          - PROCESS step  → WorkflowProcess.execute()
          - PARTICIPANT step → inbox task for user/group
          - DYNAMIC_PARTICIPANT → ParticipantStepChooser.getParticipant()
          - OR_SPLIT → route expression evaluates to true/false
          - AND_SPLIT → parallel branches, AND_JOIN waits for all

---

Common Task Patterns (End-to-End)

Pattern A: New custom approval workflow

1. Load workflow-model-design + workflow-development sub-skills
2. Design model: START → PARTICIPANT (reviewer) → PROCESS (approve/reject logic) → END
3. Implement WorkflowProcess for the approve/reject step
4. Deploy model XML to /conf/global/settings/workflow/models/
5. Deploy OSGi bundle with the process step

Pattern B: Auto-process content on upload

1. Load workflow-launchers sub-skill
2. Configure a cq:WorkflowLauncher for NODE_ADDED on the DAM path
3. Point it to your workflow model at /var/workflow/models/
4. Deploy launcher config to /conf/global/settings/workflow/launcher/config/

Pattern C: Programmatically batch-start workflows

1. Load workflow-triggering sub-skill
2. Implement WorkflowStarterService using ResourceResolverFactory + WorkflowSession
3. Map sub-service workflow-starter to workflow-process-service
4. Deploy and trigger from a Sling Scheduler or Servlet

Pattern D: "Workflow errors on host X for the past 4 hours"

1. Load workflow-triaging → classify as workflow_fails_or_shows_error
2. Suggest Splunk / Cloud Manager log search for Error executing workflow step on host + time range
3. If errors found, load workflow-debugging → map to runbook, walk decision tree
4. Return: symptom_id, runbook, evidence, remediation

Pattern E: "Workflow stuck — not advancing"

1. Load workflow-debugging → classify as workflow_stuck_not_progressing
2. Follow decision tree: check for work item → step type → specific checks
3. If thread pool suspected, guide thread dump analysis (Developer Console)
4. Return: root cause, config fix (via Git), remediation steps

---

References in This Skill

references/workflow-foundation/architecture-overview.md
references/workflow-foundation/api-reference.md
references/workflow-foundation/jcr-paths-reference.md
references/workflow-foundation/cloud-service-guardrails.md
references/workflow-foundation/quick-start-guide.md

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: adobe
• Source: adobe/skills
• License: Apache-2.0
• Homepage: https://www.adobe.com/ai/overview.html

Install and usage instructions live in the source repository linked above.