Alert RCA with Autopilot and Email

This template runs Autopilot on a New Relic alert and sends the root cause analysis as a formatted HTML email. If Autopilot returns no usable response, the workflow sends an error notice instead. For shared prerequisites and configuration notes, see Autopilot RCA workflow templates.

Before using this workflow, ensure you have:

An Email destination configured in New Relic — see Create destinations
The destination's UUID

This workflow completes the following steps:

Runs Autopilot with a root cause analysis prompt scoped to the alert issue ID
Checks whether Autopilot returned a usable response
If Autopilot succeeds, formats the response as HTML and sends it to the email destination
If Autopilot fails, sends an error email with the issue ID

This workflow uses newrelic.autopilot.run and newrelic.notification.sendEmail.

name: autopilotEmailRCA
description: Automatically analyze New Relic alerts with Autopilot and send a formatted incident analysis report via email
workflowInputs:
  issueId:
    type: String
    required: true
    validations:
      - errorMessage: Must be a valid New Relic issue ID in UUID format
        pattern: '^[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}$'
        type: regex
  emailDestinationId:
    type: String
    required: true
    validations:
      - pattern: ^[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}$
        errorMessage: "Invalid format. Please provide a valid UUID."
        type: regex
steps:
  - name: runAutopilot
    type: action
    action: newrelic.autopilot.run
    version: 1
    inputs:
      prompt: >-
        Investigate New Relic issue ${{ .workflowInputs.issueId }}.
        Produce an incident analysis report covering: what triggered the alert, likely root cause, affected entities, supporting evidence, and recommended next steps.
        IMPORTANT: Output your final answer DIRECTLY as clean, basic HTML. Use <h2> for section headers, <b> for bold text, <p> for paragraphs, <code> for logs/IDs, and standard <table>/<tr>/<td> tags with simple borders for tabular data. Do NOT use markdown.
      selectors:
        - name: agentResponse
          expression: .response.finalAnswer
        - name: agentResponseBlocks
          expression: '(.response.finalAnswer // "") as $r | (if ($r|type)=="string" then ($r | try fromjson catch null) else $r end) as $json | if ($json|type)=="object" and ($json.card.body|type)=="array" then $json.card.body elif ($json|type)=="array" then $json elif ($json|type)=="object" and ($json.text|type)=="string" then [{"type":"Text","value":$json.text}] else [{"type":"Text","value":($r|tostring)}] end'
    ignoreErrors: true

  - name: checkAutopilotResponse
    type: switch
    switch:
      - condition: ${{ (.steps.runAutopilot.outputs.agentResponse // "") != "" }}
        next: sendRcaEmail
    next: sendAutopilotError

  - name: sendRcaEmail
    type: action
    action: newrelic.notification.sendEmail
    version: 1
    inputs:
      destinationId: ${{ .workflowInputs.emailDestinationId }}
      subject: Autopilot RCA Report - Issue ${{ .workflowInputs.issueId }}
      message: '${{ def sanitizeHtml: gsub("(?i)<script\\b[^>]*>[\\s\\S]*?</\\s*script\\s*>"; "") | gsub("(?i)<\\s*script\\b[^>]*/?\\s*>"; "") | gsub("(?i)<style\\b[^>]*>[\\s\\S]*?</\\s*style\\s*>"; "") | gsub("(?i)<\\s*(iframe|object|embed|link|meta|form|svg)\\b[^>]*/?\\s*>"; "") | gsub("(?i)</\\s*(iframe|object|embed|form|svg)\\s*>"; "") | gsub("(?i)\\son\\w+\\s*=\\s*[^\\s>]+"; "") | gsub("(?i)javascript\\s*:"; ""); (.steps.runAutopilot.outputs.agentResponseBlocks // []) as $blocks | [ $blocks[] | if .type == "CodeBlock" then "<pre><code>" + ((.code // "") | gsub("&";"&amp;") | gsub("<";"&lt;") | gsub(">";"&gt;")) + "</code></pre>" else ((.value // "") | gsub("\\*\\*";"") | sanitizeHtml) as $clean | [ $clean / "\n" | .[] | select((. | gsub("^\\s+|\\s+$";"")) != "") ] as $lines | (reduce $lines[] as $line ([]; ($line | test("^\\s*\\|?\\s*:?-+:?\\s*(\\|\\s*:?-+:?\\s*)+\\|?\\s*$")) as $isSep | ($line | contains("|")) as $isPipe | if $isSep then . elif $isPipe then if (length > 0 and .[-1].type == "table") then .[0:-1] + [ .[-1] + {rows: (.[-1].rows + [$line])} ] else . + [ {type:"table", rows: [$line]} ] end else . + [ {type:"line", value: $line} ] end )) as $rawSegments | [ $rawSegments[] | if .type == "table" and (.rows | length) < 2 then {type:"line", value: .rows[0]} else . end ] as $segments | [ $segments[] | if .type == "table" then "<table style=\"width:100%;border-collapse:collapse;font-family:Arial,sans-serif;\">" + ([ .rows | to_entries[] | . as {key: $i, value: $row} | ($row | sub("^\\s*\\|";"") | sub("\\|\\s*$";"")) as $trimmed | [ $trimmed / "|" | .[] | gsub("^\\s+|\\s+$";"") ] as $cells | (if $i == 0 then "th" else "td" end) as $tag | (if $i == 0 then "text-align:left;padding:8px;border-bottom:2px solid #ccc;" else "padding:8px;border-bottom:1px solid #eee;" end) as $style | "<tr>" + ([ $cells[] | "<" + $tag + " style=\"" + $style + "\">" + . + "</" + $tag + ">" ] | join("")) + "</tr>" ] | join("")) + "</table>" else .value as $line | ($line | sub("^\\s+";"")) as $t | if ($t | startswith("<")) then $line elif ($line | startswith("### ")) then "<h3>" + ($line | ltrimstr("### ")) + "</h3>" elif ($line | startswith("## ")) then "<h2>" + ($line | ltrimstr("## ")) + "</h2>" elif ($line | startswith("# ")) then "<h2>" + ($line | ltrimstr("# ")) + "</h2>" elif ($line | startswith("- ")) then "<p>&#8226; " + ($line | ltrimstr("- ")) + "</p>" else "<p>" + $line + "</p>" end end ] | join("\n") end ] | join("\n") }}'
    next: end

  - name: sendAutopilotError
    type: action
    action: newrelic.notification.sendEmail
    version: 1
    inputs:
      destinationId: ${{ .workflowInputs.emailDestinationId }}
      subject: Autopilot RCA Failed - Issue ${{ .workflowInputs.issueId }}
      message: >-
        Autopilot could not produce findings for issue ${{ .workflowInputs.issueId }}. Please investigate manually.
    next: end

How it works

The runAutopilot step uses two selectors to handle finalAnswer shape variability at the data-extraction layer:

agentResponse captures the raw finalAnswer value. The checkAutopilotResponse step uses this to determine whether Autopilot returned anything.
agentResponseBlocks normalizes the value into an array of typed block objects, handling three possible finalAnswer shapes: a plain string, a JSON-encoded object with a .card.body array, and a bare array. This gives the sendRcaEmail step a consistent structure to iterate over regardless of how Autopilot formatted its response.

sendRcaEmail processes the blocks inline:

CodeBlock items are rendered as <pre><code> with HTML entities escaped.
Other blocks pass through sanitizeHtml, which strips <script>, <style>, inline event handlers, and other potentially unsafe markup.
Markdown-style pipe tables are detected and converted to <table> elements with basic inline styles.
Heading prefixes (# , ## , ### ) and bullet markers (- ) are converted to the corresponding HTML tags.
All remaining lines are wrapped in <p> tags.

Unlike the Alert RCA with Autopilot and Slack template, HTML formatting is handled entirely within the sendRcaEmail step — no separate assign step is needed.

Autopilot RCA workflow templates

Shared prerequisites and configuration notes for all Autopilot RCA templates

Autopilot overview

Learn how Autopilot investigates alerts and incidents

Conditional logic

Use switch statements to branch on success or failure

Send email notifications

Full reference for the sendEmail action

Create destinations

Configure the Email destination required by this workflow

Alert RCA with Autopilot and Slack

Post RCA findings to Slack instead of email

Troubleshoot Workflow Automation

Solutions for common Autopilot and workflow integration issues