Skip to main content
⏳ Estimated read time: 5 min read

New Conversation

New Conversation

Use + New Conversation button to start a fresh thread with clear context and expected outcomes. In Aiden 2.0, better prompts lead to faster, more reliable results.

Available for roles: Admin Workspace Admin Workspace User

Scope: Workspace

You can start a new conversation when you need to:

  • Triage a new incident or alert.
  • Investigate a service failure.
  • Request automation for a workflow.
  • Run a scoped analysis for one workspace.
  • Avoid mixing unrelated tasks in an existing thread.

Prompt Structure Best Practice​

For best results, structure prompts in this order:

  1. Role and objective: Tell Aiden what role it should simulate and what success looks like.
  2. Context and source data: Include alert details, entity identifiers, timestamps, labels, and integration names.
  3. Constraints: State what tools or integrations to use or avoid.
  4. Expected output format: Specify exact sections you want in the response.
  5. Next action requirement: Ask Aiden to persist, classify, or route the result if needed.

Sample Prompts​

tip
  • Keep one objective per conversation.
  • Include exact integration names to avoid ambiguity.
  • Provide explicit output sections if the result is used by downstream systems.
  • Reuse successful prompts as templates for your team.

  • Incident triage prompt​

    You are assisting an on-call SRE. Investigate this firing alert and summarize likely root cause, impact, immediate mitigation, and recovery checks.
    Service: <service-name>
    Alert: <alert-title>
    Started at: <timestamp>
    Use only these integrations: <integration-names>
    Output sections: root cause, blast radius, impact, mitigation, verification, follow-up hardening.

  • Deployment failure prompt​

    Act as a release engineer. A deployment failed in production after the latest change.
    Compare current and previous deploy state, identify likely regression point, and propose rollback or fix-forward plan.
    Use only workspace integrations and provide exact validation steps before and after mitigation.

  • Noisy alert reduction prompt​

    Review this alert stream and identify noise patterns.
    Recommend ingestion filters, severity normalization, and auto-investigation rules that reduce alert fatigue without hiding true incidents.
    Return: findings, filter rules, risk notes, and rollout plan.

  • Post-incident hardening prompt​

    Based on this incident summary, propose hardening actions across observability, runbooks, and policy guardrails.
    Prioritize by impact and implementation effort.
    Return a 30-60-90 day action plan.

  • Advanced Example: Full On-Call SRE Prompt​

    Use this format when you need detailed triage and an explicit handoff payload.

    You are assisting an on-call SRE minimizing MTTR for a Grafana alert. Alert: "5xx alerts detected for service "consoleapi" and operation "post"". UID: sha256:c7ab3e03dc64c9e4. State: firing. Service/context: consoleapi. Active since: 2026-05-21T08:08:50.000Z. Use the same tenant integrations as the SRE war room: grafana=grafana-integration. Grafana data in this UI is from integration "grafana-integration". Prefer these integration names when calling tools, do not substitute a different Grafana or SCM instance. Labels: __alert_rule_uid__=dffw24qy938qoe, alertname=APM5xxErrors, grafana_folder=OpsVerse-Alerts, normalizedOperation=post, service=consoleapi. Annotations: __orgId__=1, __value_string__=[ var='A' labels={normalizedOperation=post, service=consoleapi} value=1 ], [ var='C' labels={normalizedOperation=post, service=consoleapi} value=1 ], __values__={"A":1,"C":1}, description=5xx alerts detected for service "consoleapi" and operation "post", summary=5xx alerts detected for service "consoleapi" and operation "post". Respond with: (1) likely root cause and blast radius, (2) customer/user impact, (3) immediate mitigation steps, (4) how to verify recovery, (5) follow-up hardening. When your analysis is complete, persist the outcome by calling the SRE Copilot app MCP tool submit_alert_triage_result (POST /mcp) with alert_uid from metadata, status (completed, partial, or failed), summary, root_cause, impact, mitigation, verification, alert_category (needs_attention, validated, needs_review, false_positive), and alert_title / alert_severity from the alert context.
Last updated on