Scoping, architecting, building and debugging automations in Make.com and n8n, plus voice agent prompts, SMS flows, webhooks and structured JSON output. Hard rules go in the workflow logic, not in the prompt. 12 prompts, free to copy.
Replace each {{VARIABLE}} with your own
details. Works with ChatGPT, Claude and Gemini.
AUT-001
Automation discovery
Use when: Scoping a new automation before building it.
Variables{{PROCESS_DESCRIPTION}}
Run automation discovery on this process: {{PROCESS_DESCRIPTION}}.
Document the current manual process exactly as it happens, including
exceptions. Identify the trigger, the inputs, the decision points, the
actions, and the outputs. Flag anything the description leaves unclear
before proposing an architecture.
Use when: Designing the automation before building it.
Variables{{PROCESS}}{{PLATFORM_PREFERENCE}}
Design the workflow architecture for: {{PROCESS}}. Platform preference:
{{PLATFORM_PREFERENCE}} (or "no preference — recommend one").
Recommend one architecture, state the reason in a line, mention an
alternative only if it differs materially. Use deterministic rules
(routers, filters, state) for any hard business rule — never rely on a
prompt instruction alone to enforce a sequencing rule like "don't book
before the quote is accepted." Define error handling and human escalation
conditions.
Use when: Building a specific automation in Make.com.
Variables{{TASK}}{{CURRENT_STATE}}
Give the exact Make.com build for: {{TASK}}. Current scenario state:
{{CURRENT_STATE}}.
For each module: module name (only where confirmed current — otherwise
describe by function), whether it's a trigger/action/search, exact field
mappings, whether each value is a mapped pill or typed text, any filter
condition needed, and error handling for anything client-facing. Give a
test input and the expected result.
Give the exact n8n build for: {{TASK}}. Current workflow state:
{{CURRENT_STATE}}.
For each node: node type, exact configuration, expressions used (with
`{{ }}` syntax), credential requirements, and error-workflow attachment
for anything client-facing. Give a test input and the expected result.
Use when: A Make scenario is failing or misbehaving.
Variables{{SYMPTOM}}{{SCREENSHOT_OR_CONFIG}}
Debug this Make scenario. Symptom: {{SYMPTOM}}. Evidence: {{SCREENSHOT_OR_CONFIG}}.
Work through: last confirmed working step, first failing step, input and
output at the failing step, field names and types, authentication, time
zone, routing conditions, empty values, duplicate or stale data. Read only
what's actually shown in any supplied evidence — never claim to see a
value or module number that isn't visible. Give the next exact click or
check.
Use when: An n8n workflow is failing or misbehaving.
Variables{{SYMPTOM}}{{EXECUTION_DATA}}
Debug this n8n workflow. Symptom: {{SYMPTOM}}. Execution data:
{{EXECUTION_DATA}}.
Same debugging sequence as the Make prompt, applied to n8n's node-level
input/output inspection. Check credentials and expressions specifically,
since these are common n8n-specific failure points.
Write the Retell system prompt for {{BUSINESS}}.
Objective: {{OBJECTIVE}}
Data to collect: {{DATA_TO_COLLECT}}
Prohibited: {{PROHIBITED_ACTIONS}}
Escalate on: {{ESCALATION_CONDITIONS}}
Cover role, business, objective, permitted actions, prohibited actions
(explicit — never let the agent promise a price, an availability slot, or
a diagnosis unless specifically permitted), conversation stages, data
collection order, confirmation behaviour, escalation, and failure handling
for unclear speech. Deliver the complete prompt text, not a description.
Build the ClickSend {{FLOW_TYPE}} flow (inbound/outbound/two-way) on
{{PLATFORM}} (Make/n8n).
Cover: international number format, message length and Unicode
considerations, dedicated number versus sender ID requirements for
two-way, opt-out handling, and delivery status if needed. Never include a
real API key — use a placeholder.
Use when: Wiring a WordPress form to an automation.
Variables{{FORM_PLUGIN}}{{TARGET_WEBHOOK}}
Wire {{FORM_PLUGIN}} to {{TARGET_WEBHOOK}}.
Give the exact method (plugin webhook setting, or the `wpcf7_before_send_mail`
hook approach for Contact Form 7 without a plugin). State that field names
must be confirmed from the actual form, not assumed. Flag spam protection
as a prerequisite before wiring to anything that sends an SMS or creates a
record per submission.
Use when: Automating a Gmail send, search or reply.
Variables{{ACTION}}{{PLATFORM}}
Build the Gmail {{ACTION}} (send/search/reply) automation on {{PLATFORM}}.
For a reply, use the thread or message ID for threading, not subject-line
matching. State the required OAuth scope. Never place a credential in the
output.
Use when: Automating a Calendar event or task creation.
Variables{{PURPOSE}}{{PLATFORM}}
Build the Google Calendar automation for: {{PURPOSE}}, on {{PLATFORM}}.
Clarify Event versus Google Task if the purpose is ambiguous. Include the
explicit time zone, the calendar ID, and — for anything client-facing —
the full relevant context in the event description. Consider duplicate
prevention if this could be triggered by a retried webhook.
Use when: Getting reliable structured output from an AI step.
Variables{{FIELDS_NEEDED}}
Define a structured output schema for: {{FIELDS_NEEDED}}.
Give a valid JSON schema, correctly typed, with required fields marked.
Include a fallback instruction for the AI step (state it does not know
rather than guess) and a parse-failure handling recommendation for the
receiving automation.