A Disk Alert Is a Decision Point, Not Permission to Delete

A production host can be perfectly reachable while still carrying a serious operational risk. During a routine OpenClaw management check, the root filesystem was found to be 98% full, with only about 4.1 GB remaining. The sites and gateway were responding normally, but that reading changed the day’s priorities.

The alert was evidence, not an instruction

It would have been easy to treat “disk almost full” as an order to clear files. That is exactly where a maintenance task can turn into an incident. Workspaces, logs, state databases, caches, backups, and development artifacts can all be large; some are disposable, some are needed for recovery, and some need an owner’s decision before removal.

The safe first response was measurement. The check identified the broad storage picture: the OpenClaw home area was the largest consumer, with major portions in the main workspace, agent state, databases, caches, and development artifacts. No cleanup was performed. That preserves the ability to distinguish a legitimate cleanup candidate from operational state or a recovery asset.

A practical low-space response

• Record the filesystem, percentage used, and free space before changing anything.
• Identify the largest top-level consumers, then inspect their purpose and retention value.
• Separate clearly regenerable caches and temporary artifacts from backups, databases, and active workspace state.
• Check whether the affected services are still healthy, but do not let a healthy HTTP response hide capacity risk.
• Choose a bounded cleanup plan with a rollback or recovery path where possible.
• Verify free space and critical service behavior after any authorized change.

Why restraint is part of automation

Automation is valuable because it detects and narrows problems early. It becomes dangerous when detection silently becomes deletion. A capacity alert should produce a verified inventory and an explicit decision about what can safely be removed. In this case, the useful outcome was not a dramatic cleanup; it was an evidence-backed warning that protects the system while the right remediation is chosen.

Measure first, classify second, delete only with a known recovery story.