engineeringoperations

Automation Workflow Design

Design and document automation workflows for n8n, Make, Zapier, and Power Automate. Covers trigger-action chain mapping, conditional branching, error handling, retry logic, and workflow diagrams with platform-specific node references.

automationworkflown8nmakezapierpower-automateintegrationno-code

Works well with agents

Integration Engineer Agent No-Code Builder Agent Workflow Automator Agent

Works well with skills

Integration Specification Runbook Writing

$ npx skills add The-AI-Directory-Company/(…) --skill automation-workflow-design

automation-workflow-design/

SKILL.md

Markdown

1
2	# Automation Workflow Design
3
4	## Before you start
5
6	Gather the following from the user:
7
8	1. What is the trigger? (Webhook, schedule, database event, form submission, file upload)
9	2. What is the desired outcome? (Record created, notification sent, file transformed, data synced)
10	3. Which platform? (n8n, Make, Zapier, Power Automate, or platform-agnostic)
11	4. What systems are involved? (APIs, databases, SaaS tools, file storage)
12	5. What is the expected volume? (Runs per hour/day, payload sizes)
13	6. What happens on failure? (Retry, alert, fallback, manual queue)
14
15	If the user says "automate this process," push back: "Which trigger starts it, what systems are involved, and what should happen when a step fails?"
16
17	## Workflow mapping procedure
18
19	### Step 1: Define the trigger
20
21	Identify the event that starts the workflow. Document:
22
23	- Trigger type: Webhook, polling, cron schedule, event subscription
24	- Payload shape: Fields available from the trigger (provide a sample JSON)
25	- Frequency: Expected invocations per time window
26	- Idempotency: Whether duplicate triggers are possible and how to handle them
27
28	### Step 2: Map the action chain
29
30	For each step after the trigger, document in sequence:
31
32	```
33	Step [N]: [Action name]
34	Platform node: [Specific node/module name for the target platform]
35	Input: [Which fields from previous steps]
36	Transform: [Any data mapping or formatting]
37	Output: [Fields produced for downstream steps]
38	Error behavior: [Retry N times / skip / halt workflow]
39	```
40
41	Group related steps into logical stages: Intake, Processing, Output, Notification.
42
43	### Step 3: Add conditional branches
44
45	For each decision point:
46
47	- State the condition as a boolean expression using available fields
48	- Document the true-path and false-path step sequences
49	- Identify whether branches rejoin or terminate independently
50
51	### Step 4: Design error handling
52
53	For every step that calls an external service:
54
55	- Retry strategy: Count, backoff interval, max wait
56	- Timeout: Maximum seconds before treating as failure
57	- Dead letter handling: Where failed items go (error queue, log table, notification)
58	- Circuit breaker: If failure rate exceeds threshold, pause workflow and alert
59
60	### Step 5: Draw the workflow diagram
61
62	Produce a text-based diagram showing the flow:
63
64	```
65	[Trigger: New form submission]
66	\|
67	v
68	[Validate payload] --invalid--> [Log error + notify]
69	\|valid
70	v
71	[Enrich: Fetch customer from CRM]
72	\|
73	v
74	[Branch: Is enterprise?]
75	\|yes \|no
76	v v
77	[Create Jira ticket] [Add to Mailchimp list]
78	\| \|
79	v v
80	[Notify #sales] [Send welcome email]
81	```
82
83	Include error paths as dashed or labeled branches.
84
85	## Platform-specific notes
86
87	n8n: Use Error Trigger node for global error handling. Reference nodes by exact name (HTTP Request, IF, Switch, Merge). Use expressions with `{{ $json.field }}` syntax.
88
89	Make (Integromat): Use error handler routes (Break, Resume, Ignore, Rollback). Reference modules by app name + action. Note the 15-minute default scenario timeout.
90
91	Zapier: Use Paths for conditional logic. Note that error handling requires Formatter + Filter workarounds or custom code steps. Multi-step zaps have a 30-second per-step timeout.
92
93	Power Automate: Use Scope actions for try-catch patterns. Configure run-after settings (succeeded, failed, skipped, timed out) on each action. Use `@outputs('step')['statusCode']` for error inspection.
94
95	## Workflow document template
96
97	```markdown
98	# Workflow: [Name]
99
100	## Overview
101	- Trigger: [Event description]
102	- Platform: [n8n / Make / Zapier / Power Automate]
103	- Schedule: [If applicable]
104	- Owner: [Team or person]
105
106	## Steps
107	\| # \| Action \| Node/Module \| Input \| Output \| Error Handling \|
108	\|---\|--------\|-------------\|-------\|--------\|----------------\|
109	\| 1 \| ... \| ... \| ... \| ... \| ... \|
110
111	## Branches
112	\| Condition \| True path \| False path \|
113	\|-----------\|-----------\|------------\|
114
115	## Error handling
116	- Global error handler: [Description]
117	- Dead letter destination: [Queue / table / channel]
118	- Alert channel: [Slack / email / PagerDuty]
119
120	## Volume & limits
121	- Expected runs: [N per hour/day]
122	- Rate limits: [Per-API limits to respect]
123	- Payload size limits: [If applicable]
124	```
125
126	## Quality checklist
127
128	Before delivering the workflow design, verify:
129
130	- [ ] Every step has explicit input fields, output fields, and error behavior
131	- [ ] Conditional branches document both paths and specify whether they rejoin
132	- [ ] Error handling covers retries, timeouts, dead letters, and alerting
133	- [ ] The workflow diagram matches the step table exactly
134	- [ ] Platform-specific node names and syntax are correct for the target platform
135	- [ ] Idempotency is addressed for the trigger and any create/update operations
136	- [ ] Volume estimates and rate limits are documented
137
138	## Common mistakes
139
140	- No error handling on HTTP steps. Every external call can fail. Skipping retry and timeout config means silent data loss in production.
141	- Ignoring idempotency. Webhooks can fire twice. Without deduplication (check by ID before create), you get duplicate records.
142	- Hardcoding credentials in workflow steps. Use the platform's credential store or environment variables. Never embed API keys in node configurations.
143	- Linear-only thinking. Real workflows branch. If you have zero conditional nodes, you probably have not considered validation failures or edge cases.
144	- Missing the "what if the whole workflow fails" path. Individual step retries are not enough. Design a global error handler that captures the workflow run ID, failed step, and payload for manual replay.
145	- Not documenting rate limits. A workflow that runs fine at 10/hour breaks at 1000/hour when a batch import triggers it. State the ceiling.
146