| name | os-health |
| description | Inspect Mac host speed risks and Agentic OS generated-artifact hygiene with observe-only reports and approval-gated cleanup classes. |
OS Health
Use when the user asks to investigate a sluggish Mac, heavy indexing, OrbStack or
runtime pressure, stale worktrees, or Agentic OS-generated logs, runs, artifacts,
and worklogs.
Canonical program: harness/shared_factory/00-programs/host_agentic_os_health/.
Load
- Route to the installed Agentic OS root and then to the owning project when a
work item or worktree registry is named.
- Read
harness/shared_factory/00-programs/host_agentic_os_health/program.md
when present.
- Read
harness/commands/os-health.md for the command contract.
- Read
harness/commands/system-tool-registry.md before non-trivial host
inspection or cleanup.
Workflow
-
Run an observe-only report:
agentic-os health report --root <os-root> --output <work-item>/artifacts/health-report.yml
-
Review host pressure:
- disk and memory
- Spotlight/indexing status
- hot processes
- OrbStack status when available
-
Review Agentic OS pressure:
- generated directories and sizes
- registered worktrees and git state
- active vs complete work-item locations
-
Separate follow-up actions:
safe-auto: allowed only by an approved policyapproval-required: ask before mutationmanual-only: user/operator handles directly
-
Use narrower cleanup commands only after their own dry-run report is reviewed.
Safety Rules
- Never delete files from this skill without explicit approval.
- Never stop or restart OrbStack, Docker, dev servers, launch agents, or runtime
supervisors without explicit approval.
- Never disable Spotlight globally.
- Never remove dirty or untracked worktrees without path-specific approval.
- Keep local paths in internal receipts only.
Output Contract
Report:
- report command and artifact path
- host bottleneck candidates
- generated artifact totals
- worktree totals
- approval-required actions
- skipped or unavailable checks
