Kafka Topic Audit

Kafka Topic Configuration Audit

Audits all topic configurations against production best practices. Misconfigured topics are the #1 cause of Kafka data loss - engineers create topics and forget to tune them.

Target environment: $ARGUMENTS

Workflow

Copy this checklist and track your progress:

Audit Progress:
- [ ] Step 1: Check environment health
- [ ] Step 2: Fetch all topics
- [ ] Step 3: Audit configurations against best practices
- [ ] Step 4: Check metadata completeness
- [ ] Step 5: Detect orphaned topics
- [ ] Step 6: Run consistency checks
- [ ] Step 7: Generate report

1. Check environment health for a high-level summary
2. Fetch all topics and their configurations
3. Audit each topic against best practices (see references/audit-rules.md)
4. Cross-reference metadata for completeness
5. Detect orphaned topics with no consumers
6. Report findings with prioritised recommendations

Step 1: Environment Overview

Use the Lenses MCP check_environment_health tool to get a quick summary:

• Broker count, topic count, consumer count, connector count
• Any existing issues flagged by Lenses

Expected output: Environment health summary with broker, topic and consumer counts.

Validation: If the environment is unhealthy or unreachable, stop and report the connection issue before proceeding.

Step 2: Fetch All Topics

Use the Lenses MCP list_topics tool to retrieve all topics with their configurations in one call.

For topics that need deeper inspection, use:

• get_topic for detailed config including partitions and consumers
• get_topic_broker_configs for broker-level config overrides
• get_topic_partitions for partition-level message counts and bytes

Expected output: Full list of topics with their configurations. If zero topics are returned, report this and stop.

Step 3: Audit Configurations

For each topic, check against the thresholds in references/audit-rules.md:

• Replication factor - RF=1 is critical, RF=2 is a warning in production
• Retention policies - unbounded growth, too short or excessively long
• Partition count - single-partition bottlenecks or excessive partitions
• Compaction settings - compact without keys, delete for state topics
• Naming conventions - must follow {domain}.{entity}.{event} pattern

Step 4: Metadata Completeness

Use the Lenses MCP list_topic_metadata tool to check:

• Topics missing descriptions
• Topics missing tags
• Topics without registered schemas (key or value)

Use list_datasets with filters (is_compacted, has_records) to find anomalies.

Step 5: Orphan Detection

For each topic, use list_consumer_groups_by_topic to check for active consumers.

• Warning: Topics with zero consumer groups (may be orphaned/dead)
• Suggestion: Topics with only inactive/empty consumer groups

Step 6: Consistency Checks

• Flag topics in the same domain with different retention policies
• Flag topics in the same domain with different replication factors
• Flag topics with inconsistent serialisation formats within a domain

Success Criteria

Quantitative

• Triggers on 90% of topic-related queries (test with 10-20 varied phrasings)
• Completes full audit in under 15 MCP tool calls
• 0 failed MCP calls per run

Qualitative

• Report is actionable without follow-up questions from the user
• Consistent severity categorisation (critical/warning/suggestion) across runs
• Every finding includes a concrete remediation step

Examples

Example 1: Routine weekly audit

User says: "Run a topic audit on the staging environment"

Actions:

1. Check staging environment health via Lenses MCP
2. Fetch all topics and configs
3. Audit each topic against rules in references/audit-rules.md
4. Check metadata completeness and orphaned topics

Result: Full audit report with prioritised findings

Example 2: Pre-deployment check

User says: "Check if my topic configs are production-ready"

Actions:

1. Audit all topics for RF < 3, unbounded retention, single partitions
2. Flag any critical issues that would block a production deployment

Result: Report highlighting critical issues that must be fixed before go-live

Example 3: Investigate a specific topic

User says: "Is the orders.payment.completed topic configured correctly?"

Actions:

1. Fetch detailed config for the specific topic using get_topic
2. Check broker-level overrides with get_topic_broker_configs
3. Verify metadata and consumer groups

Result: Focused report on a single topic with all findings

Troubleshooting

Lenses MCP connection failed

Cause: Environment name is incorrect or Lenses agent is offline. Solution: Run check_environment_health first. Verify the environment name matches what list_environments returns.

No topics returned

Cause: Environment exists but has no topics or permissions are restricted. Solution: Confirm the cluster has topics via the Lenses UI. Check that the Lenses agent has read access.

Metadata endpoint returns empty

Cause: Schema Registry is not configured or topics have no registered schemas. Solution: This is a valid finding - report it as missing metadata rather than treating it as an error.

Output Format

## Topic Audit Report

### Environment: {name}
- Brokers: X | Topics: Y | Consumer groups: Z

### Critical (must fix)
- [topic-name] Description of the issue
  Current: {current value} | Recommended: {recommended value}

### Warning (should fix)
- [topic-name] Description of the issue
  Current: {current value} | Recommended: {recommended value}

### Suggestion (consider improving)
- [topic-name] Description of the issue
  Recommendation: How to fix it

### Summary
- X critical issues found
- Y warnings found
- Z suggestions found
- Topics audited: N
- Orphaned topics: M

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: lensesio
• Source: lensesio/agentic-engineering-for-apache-kafka
• License: MIT
• Homepage: https://lenses.io

Install and usage instructions live in the source repository linked above.