@mochi-replan, Partial Queue Adjustment

Adjust the current plan without starting from scratch. Faster and cheaper than @mochi-plan.

Context to Read

1. get_plan(), the whole current queue, including tasks already done (done: true) in the last 2 hours; skip those when deciding what is still scheduled
2. get_watchlist(), active watch items (check for overdue items that need scheduling)
3. read_mochi_file({ which: "activity" }), recent activity (last 5-10 entries)
4. The planner_notes from the queue, contains memory from the last full plan (and silent_until when the user asked for quiet)

Do NOT re-read the calendar. That is only for mochi-plan.

Two Modes

Automatic Mode (triggered by poller when needs_replan is true)

Detect what changed and adjust:

• User dragged the pet (activity shows drag/interrupted_by_user): → Remove the next scheduled move task → Add a playful reaction: { type: "notify", action: { summary: "Hey! I was going somewhere!" }, category: "reaction" } → Optionally add a new move back to a natural position

• User sent a message referencing the plan (activity shows recent chat): → Check if any queued tasks conflict with what the user said → Adjust accordingly

• Watched item status changed: → If a watch result notification came in (activity log), update planner_notes.watching_last_status → Note: watch checks are driven by the queue poller (based on nextCheckAfter), not scheduled as queue tasks. Do NOT schedule check_watching tasks.

• Quiet requested (planner_notes.silent_until is set and in the future): → Remove upcoming notify tasks before that time, keep only moves → When silent_until is null or past, nothing to remove

• General drift (time passed, some tasks executed): → Rebalance remaining tasks if timing feels off

Explicit Mode (triggered by chat agent with user request)

The chat agent will provide context about what the user wants. Common requests: