Deployment & Operations

CI/CD pipelines are configuration-heavy and well-suited for agents. Monitoring and maintenance are ongoing concerns where agents help with setup, incident response, and routine tasks.

CI/CD & Deployment

Where agents help

Pipeline configuration: GitHub Actions, GitLab CI, Jenkins, CircleCI. Agents know common patterns and adapt to your stack.

Infrastructure as Code: Terraform, CloudFormation, Pulumi—generate resources, modify existing infrastructure, create reusable modules.

Deployment automation: Dockerfiles, Kubernetes manifests, Helm charts, deploy scripts, rollback procedures.

Environment configuration: Environment variables, secret management, configuration files.

What to watch for

Security — CI/CD is security-critical. Agent-generated pipelines may expose secrets in logs, use overly permissive permissions, or skip security scanning. Always security-review CI/CD changes.

Vendor quirks — Each CI system has differences. Test generated pipelines thoroughly.

Stateful resources — Infrastructure changes can cause deletion, downtime, cost surprises, or data loss. Use plan/apply patterns.

Prompt patterns

Pipeline generation:

Create a GitHub Actions workflow for a [language/framework] project.

Requirements:
- Run tests on PR
- Deploy to [environment] on merge to main
- Use [specific services/tools]

Follow security best practices.

Infrastructure:

Generate Terraform for [resource type] with:
- [Specific requirements]
- [Constraints]

Follow the patterns in [existing file/module].

Dockerfile:

Create a Dockerfile for [application type].

Requirements:
- Multi-stage build
- Minimal final image
- Run as non-root
- [Other requirements]

Monitoring & Maintenance

Monitoring setup

Alert configuration: Prometheus rules, CloudWatch alarms, Datadog monitors. Agents understand common patterns for what to monitor.

Dashboard creation: Grafana dashboards, Kibana visualizations. Describe what you want to see; get a starting configuration.

Log aggregation: Parsing rules, search queries, anomaly detection.

Incident response

Root cause analysis: “Here’s the error and recent changes. What could cause this? What should I check?”

Debugging assistance: Analyze logs, interpret stack traces, trace request flows, identify error patterns.

Post-mortem drafting: “Help me write a post-mortem for [incident]. Include: summary, timeline, root cause, impact, remediation, lessons learned.”

Maintenance tasks

• Dependency updates — Analyze versions, identify breaking changes, generate update PRs
• Technical debt cleanup — Identify improvement patterns, generate refactoring plans
• Performance optimization — Analyze issues, suggest approaches, generate benchmarks
• Security maintenance — Vulnerability remediation, patch application, configuration hardening

What to watch for

• Alert fatigue — Generated alerts may be too sensitive or poorly calibrated. Tune based on real experience.
• Dashboard overload — More dashboards isn’t better. Ask: What decision does each panel inform?
• Maintenance scope creep — Bound tasks to the actual problem being solved.

Prompt patterns

Alert rules:

Create Prometheus alerting rules for a [service type].

Monitor:
- Error rate exceeding [threshold]
- Latency above [threshold]
- Resource utilization above [threshold]

Include appropriate severity levels and annotations.

Incident investigation:

I'm seeing [error/symptom] in production.

Relevant context:
- [Recent changes]
- [Error logs]
- [Metrics]

What could cause this? What should I check first?

Runbook:

Create a runbook for handling [type of incident].

Include:
- Detection (how we know it's happening)
- Triage (how to assess severity)
- Mitigation (immediate actions)
- Resolution (full fix)
- Follow-up (post-incident)