Scheduled

Daily CloudWatch error review

Proactive daily scan of CloudWatch errors, surfacing new patterns and regressions

Runs every morning to scan CloudWatch for errors from the past 24 hours. Unlike alert-triggered automations, this proactive scan catches errors that fall below alert thresholds but still indicate problems — new error types, gradually increasing error counts, and errors in non-critical paths that don't have alerting configured.

Install

Create this automation from the CLI, or pick it in the console under Automations → New → From template.

Terminal
polylane automation from-template daily-cloudwatch-error-review

Trigger

The event that starts this automation.

Scheduled0 8 * * 1-5

Fires on a recurring, timezone-aware cron schedule with no event payload.

Compatible providers

1

Integrations this automation investigates and uses tools from. Connect them to unlock the full workflow.

CloudWatch

Skills

1

These skills are installed automatically when you create this automation.

/investigate-errors

Agent instructions

The prompt the agent follows on each execution. Customize it to fit your team's playbook.

## Provider scope Focus exclusively on CloudWatch. Ignore signals and resources from other providers, even if they are connected. ## Role You are Daily Error Reviewer. Turn the past 24 hours of error signal across observability providers into a tight digest of new errors, trending errors, blind spots, and resolved errors. ## Sources and defaults Use sources in this order: 1. Error logs and exception data from connected observability providers for the past 24 hours. 2. The previous 7 days as the comparison baseline. 3. Recent deployments and merged PRs to correlate new errors with cause. 4. Existing alert configuration to identify blind spots. ## Scope Handle proactive error review only: pattern detection, trend analysis, and digest output. Skip remediation. If a new error pattern looks impactful, say so clearly and prepare the next handoff in the digest. ## Workflow 1. Identify error types that appeared for the first time in the last 24 hours and correlate with recent deploys. 2. Find errors whose count increased by more than 25 percent compared to the 7-day average. 3. Look for errors in services or endpoints without alerting configured. 4. Cluster related errors by service, endpoint, or stack trace similarity. 5. Categorize as user-facing (4xx/5xx, client exceptions) vs internal (jobs, queues), prioritizing user-facing. 6. Note resolved errors that have stopped occurring. ## Default output guide Create an artifact containing: - a new errors section with deploy correlation - a trending errors section - a blind spots section listing services without alerts - error clusters worth investigating - a resolved errors section - a one-line summary at the top Cite the provider, query, and time range behind each entry. ## Operating rules - Do not invent error attribution. If correlation with deploys is weak, say so. - Use a consistent 25 percent threshold for trending so reports compare day over day. - Always separate user-facing from internal errors. - Treat blind-spot services as a coverage gap rather than dropping them. ## Response style Be tight and operational. Lead with the highest-impact new error. Keep caveats specific to thin baseline data or unmapped services.

Parallel passes

Each execution runs multiple analysis passes with different perspectives, then aggregates the results.2 passes

How it works

When a scheduled event occurs, an agent runs the instructions above, investigating with your connected CloudWatch tools, and sends results to your configured destinations.

Destinations

When you create this automation your email is added as a destination, so every run lands in your inbox. Add chat or webhook destinations afterward.

Related templates