Investigate CRM Production Incidents with an AWS Agent Skill

Where this fits

You are the on-call engineer for the CRM. An alert fires: SQS backlog for the customer-sync queue is climbing. The webhook payload gives you the incident title, but no context. Instead of jumping between CloudWatch and Lambda logs to reconstruct what "normal" looks like, an agent using this skill knows the CRM's topology and the exact AWS-native toolchain to investigate. It guides whoever triages the alert—human or AI—through a disciplined root-cause path.

Why agents benefit

• Distinguishes incident types by exact webhook title, so an agent never conflates a queue backlog with database latency.
• Restricts investigation scope to relevant AWS services (REST API, SQS, Lambda, Aurora DSQL), preventing exploration drift.
• Recommends concrete observability actions: CloudWatch metric math, SQS depth and age checks, Lambda concurrent executions, X-Ray traces, Aurora performance insights.
• Steps through a decision tree for RCA, including isolation rules that force a fresh investigation for each new incident title.
• Encodes operational knowledge that normally lives in tribal runbooks, making it accessible to autonomous agents.

Practical scenarios

SQS Message Backlog Spike

The consumer Lambda falls behind because a malformed event keeps failing. The skill points the agent to queue metrics, dead-letter queue depth, and suspicious message IDs, then correlates with consumer error logs to identify the poison message.

RDS Read Latency/CPU Utilization High

A new report query is hammering the reader endpoint. The agent is directed to Aurora DSQL performance metrics, slow query logs, and active session analysis to pinpoint the SQL and suggest an index or query rewrite.

High Lambda Error Rate, All Invocations Failing

Every invocation fails right after a deployment. The skill guides the agent to check release history, recent log-stream errors, and X-Ray traces to determine whether a config change or code path regression is the cause.

Add it to your agent workflow

Install the skill from the AWS tools-for-devops-agent repo:

npx skills add aws/tools-for-devops-agent --skill crm-production-investigation-guidelines

Then in your agent instructions, use it when an incident payload arrives:

You are triaging a CRM production incident. Use the crm-production-investigation-guidelines skill. Start with the webhook title exactly as provided, and follow the AWS-native investigation steps until you can state a root cause.

Compare before adopting

When evaluating this skill, compare it against your existing runbooks and generic observability prompts. Check how recently the repo was updated, whether the skill's decision trees align with your CRM's architecture, and if the AWS-native tool references match the account you operate. It is a focused skill, not a platform, so consider whether your agents can already execute the recommended API calls.

Why it is worth tracking

Backed by AWS and part of the tools-for-devops-agent repository, this skill shows that practical incident triage knowledge is being packaged for agents. The 42-star count is modest, but the real signal is the structured isolation rules and concrete AWS service mapping—exactly what you would write after several painful on-call cycles. Revisit it when you are improving your own agent skills or standardizing incident workflows. The skill's page at OpenAgentSkill (https://www.openagentskill.com/skills/aws-crm-production-investigation-guidelines) includes the metadata and install details for quick comparison.