Weekly self-maintenance job: clean up stale worktrees, memory, and cron jobs

[morgan-fetch]

Idea: a recurring (weekly) job where the agent takes care of its own space — the agent workspace on this machine.

Motivation: staleness accumulates over time. Recent example: a second, duplicate chronicle cron job was created without checking the existing one — exactly the kind of drift a cleanup pass catches. Right now things are fine, but after weeks/months of use this will keep happening.

Scope of the cleanup pass:

• Stale worktrees — prune removed/merged feature worktrees under projects/, check for worktrees left on merged branches.
• Stale memory — trim MEMORY.md (keep only up-to-date facts), review/forget outdated memory-store entries, drop stale digest items.
• Stale cron jobs — flag or remove disabled, one-shot leftovers, and duplicate/overlapping schedules.
• General tidying of the workspace (chronicle, garden, temp files) as needed.

Open questions:

• Cadence: weekly feels right, but could be bi-weekly.
• Should cleanup act autonomously or propose a report for review first? (Probably report-then-act for destructive bits like deleting worktrees.)

Related: agent runtime files live in this repo (AGENTS.md, etc.), so any scheduling/automation details likely belong here too.