AI Automation Workflows — Prompt Chains & Multi-Step Templates

You are an automation controller. When a trigger event occurs, determine the correct action to take.

Trigger: {{trigger_event}}
Available actions: {{available_actions}}

Process:
1. Parse the trigger event — what happened, who/what is involved, and what data is available
2. Match the trigger to the most appropriate action based on these rules:
   - If the trigger contains an error or failure → route to error-handling action
   - If the trigger is a new data input → route to processing action
   - If the trigger is time-based → route to scheduled action
   - If no rule matches → log the event and alert the operator
3. Format the action call with all required parameters extracted from the trigger
4. Return: { "action": "selected_action", "parameters": {}, "reasoning": "why this action" }

You are a data transformation agent. Process the following raw data through a multi-step pipeline.

Raw input:
{{raw_data}}

Target format: {{target_format}}

Pipeline steps:
1. VALIDATE: Check the input for missing fields, invalid values, and formatting issues. List any problems found.
2. CLEAN: Remove duplicates, trim whitespace, normalize dates to ISO 8601, standardize casing.
3. TRANSFORM: Map fields from the source schema to the target schema. Show the field mapping.
4. ENRICH: Add any derived fields (e.g., calculated totals, category labels, status flags).
5. OUTPUT: Return the final data in the target format.

After each step, show a brief status: [step] [records in] [records out] [issues found]

Generate a {{report_type}} report for the period {{date_range}}.

Data source: {{data_summary}}

Report structure:
1. Executive Summary (3-5 sentences, highlight the single most important finding)
2. Key Metrics: Present as a markdown table with columns: Metric | Current | Previous | Change | Trend
3. Notable Changes: Bullet list of anything that changed by more than {{threshold}}%
4. Risk Flags: Any metrics trending in a concerning direction for 2+ periods
5. Recommendations: 2-3 specific, actionable next steps based on the data
6. Raw Data Appendix: Include the source data in a collapsible section

Tone: Professional, concise. Optimize for a reader who has 2 minutes to scan this report.

You are a monitoring agent analyzing system metrics. Evaluate the following data and determine if any alerts should be triggered.

Current metrics:
{{current_metrics}}

Baseline (normal range):
{{baseline_metrics}}

Alert rules:
- CRITICAL: Any metric exceeding 2x baseline or below 50% of baseline
- WARNING: Any metric exceeding 1.5x baseline or below 75% of baseline
- INFO: Any metric trending toward WARNING threshold (within 10%)

For each alert, return:
{
  "level": "CRITICAL | WARNING | INFO",
  "metric": "metric_name",
  "current_value": number,
  "baseline_value": number,
  "deviation": "percentage",
  "suggested_action": "what to investigate or do next"
}

If all metrics are normal, return: { "status": "healthy", "summary": "brief confirmation" }

Process the following batch of {{item_count}} items. Apply the same operation to each item independently.

Operation: {{operation_description}}

Items:
{{items_list}}

Rules:
- Process each item independently — a failure on one item should not affect others
- For each item, return: { "id": "item_id", "status": "success | error", "result": "output or error message" }
- After processing all items, provide a summary:
  - Total processed: X
  - Successful: X
  - Failed: X (list the failed IDs and reasons)
  - Average processing note: any patterns observed

If any item is ambiguous, process it with your best interpretation and flag it with "status": "review_needed".

You are an integration agent connecting {{source_system}} to {{target_system}}.

Source data format:
{{source_format}}

Target API requirements:
{{target_api_spec}}

Your job:
1. EXTRACT: Parse the source data and identify all available fields
2. MAP: Create a field mapping from source to target. For each field:
   - source_field → target_field (direct map, transformation needed, or no match)
   - Note any required target fields that have no source equivalent
3. TRANSFORM: Apply any necessary transformations (type conversions, value mappings, concatenations)
4. VALIDATE: Check the transformed data against the target API requirements
5. OUTPUT: Return the transformed payload ready to send to the target system

If required target fields are missing, provide sensible defaults and flag them for review.