Practical audit
These boxes are intentionally non-interactive. Use the list as a review framework and record decisions in the system where your team owns operational work.
Local developer machines
Measure AI cache, model, package, and build directories.
Identify artifacts that can be regenerated safely.
Remove abandoned project outputs after confirming ownership.
Document recurring cleanup commands for the team.
AI coding agents
Review saved sessions, transcripts, tool outputs, and scratch work.
Separate useful project knowledge from temporary execution state.
Set retention windows for logs and browser artifacts.
Remove credentials and sensitive material from retained context.
Vector databases
Record the source revision and creation date for each collection.
Detect embeddings whose source documents changed or disappeared.
Test retrieval quality before and after refreshes.
Define deletion and rebuild procedures.
Prompt systems
Assign stable versions to production prompts.
Map each prompt version to its active consumers.
Retire copied or superseded instructions.
Review saved context for conflicting guidance.
Docker and build cache
List images, volumes, layers, and caches by age and owner.
Confirm which volumes still support active services.
Prune reproducible build material on a measured schedule.
Protect stateful data from broad cleanup commands.
Model files
Inventory model names, versions, sizes, and consumers.
Remove redundant or superseded local copies.
Verify checksums and provenance for retained models.
Separate shared production assets from experiments.
Retention rules
Give every retained system an accountable owner.
Define review dates instead of relying on indefinite storage.
Document exceptions for security, legal, and debugging needs.
Measure growth and revisit the policy when system behavior changes.