edgedelta-workflows

EdgeDelta Workflows

Overview

Generate complete EdgeDelta workflow configurations — event-driven AI automation pipelines that connect monitors, external events, and schedules to AI teammates. Workflows are JSON DAGs with typed nodes (start, teammate, task, action, if-else, transform, set-state, note) executed on a distributed async engine.

Workflow Generation Process

1. Determine the Trigger

Ask what initiates the workflow:

Trigger	Start Node Config	When to Use
Manual / Monitor / Connector	"start": {}	Alert-driven, event-driven, or on-demand
Periodic (scheduled)	"start": { "periodicRun": { "schedule": { "cronExpression": "...", "timezone": "..." } } }	Recurring health checks, reports

2. Select the Template

Match the user's scenario to a template from references/templates.md:

Scenario	Template
Alert → investigate → notify	Template 1 (Monitor Alert → Email)
Scheduled check with branching	Template 2 (Daily Health Check)
Full incident response (multi-tool)	Template 3 (Jira + Slack + PagerDuty)
Code review automation	Template 4 (PR Review → Slack)
Periodic reporting	Template 5 (Capacity → Teams)
Custom / none match	Build from scratch using schema

3. Customize the Workflow

For each node, consult the appropriate reference:

• Node types and schemas: Read references/workflow-schema.md
• Action configurations: Read references/action-types.md — covers all 29 action types with complete JSON examples
• Templates: Read references/templates.md — 5 production-tested starting points

4. Validate the Configuration

Before returning the workflow, verify:

1. Exactly one start node exists
2. All node names match the pattern /^[a-zA-Z0-9_-]+$/
3. All node names are unique
4. Every node (except note) is referenced in at least one link
5. Every link references existing node names
6. No cycles in the DAG
7. If-else paths in links match the branch paths defined in the node
8. Teammate nodes have a valid agentId — built-in options: sre, devops-engineer, cloud-engineer, security-engineer, code-analyzer, issue-coordinator
9. Structured output schemas include required and match JSON Schema spec
10. Action configs use the correct config key for the action type (e.g., slackSendMessage for slack-send-message)
11. Handlebars templates use triple braces {{{ }}} for JSON/HTML content (no escaping) and double braces {{ }} for plain values
12. Integration names are user-provided — always ask if not specified

Node Quick Reference

Node	Purpose	Key Fields
start	Entry point	periodicRun? (cron + timezone)
teammate	AI agent with tools + memory	agentId, stepLevelPrompt, outputFormat, structuredOutput?
task	Direct LLM call (no tools/memory)	prompt, outputFormat, structuredOutput?
action	External integration	actionType + type-specific config (see action-types.md)
if-else	Conditional branching	branches[] with path + condition (JS expressions)
transform	Data transformation	steps[] with name + script (sandboxed JS)
set-state	Persist key-value pairs	assignments[] with key + expression
note	Documentation only	content

Data Flow

All Handlebars templates and JS expressions access three variables:

Variable	Contains	Example
data	Workflow input (trigger payload or transformed data)	{{ data.severity }}
nodes	Results from prior nodes, keyed by name	{{ nodes.investigate.summary }}
state	Values set by set-state nodes	{{ state.incidentId }}

Common Patterns

Fork after investigation — one teammate node feeds multiple actions in parallel:

investigate → create-jira
investigate → notify-slack
investigate → page-oncall

Conditional escalation — branch based on teammate output:

investigate → check-severity → (critical) → page-oncall
                             → (warning) → send-email
                             → (info) → log-only

Data normalization pipeline — clean input before processing:

start → transform (normalize) → set-state (save context) → teammate

Multi-specialist chain — sequential teammate analysis:

sre-investigate → security-review → devops-remediation → notify

Integration Checklist

When generating a workflow, confirm the user has provided (or ask for):

• Integration names for each action (e.g., Slack integration name, Jira integration name)
• Channel/project IDs for actions (Slack channel ID, Jira project ID)
• Email recipients
• PagerDuty service IDs
• Custom teammate IDs (if not using built-in agents)
• Cron schedule and timezone (if periodic)

Resources

references/

• workflow-schema.md — Complete node type schemas, link rules, trigger types, execution model, Handlebars helpers, and JS expression context. Search patterns: grep -i "node type\|schema\|field\|required" references/workflow-schema.md
• action-types.md — All 29 action type configurations with full JSON examples for Email, Slack (10), Jira (4), PagerDuty (3), Teams (3), EdgeDelta (1). Search patterns: grep -i "actionType\|slack\|jira\|pagerduty\|teams\|email" references/action-types.md
• templates.md — 5 production-tested workflow templates with customization guides. Search patterns: grep -i "template\|use case\|trigger" references/templates.md