hivehouse: cleanup

Claude Code fills a ${user_config.X} below only when a value of X was saved; only when one still reads literally, as ${user_config.X}, run hh-settings option X and use what it prints (the value saved in Claude Code's settings, else the plugin's default); a filled value is used as it is.

Run hh-resolve rules and read the file it prints first (the project's or the user's own custom/rules/RULES.md, else the plugin's). hh-cleanup is the executable that does the work; this skill is the owner's full, supervised pass. hh-cleanup is deliberately not pre-approved: it prompts every time, so the owner sees the exact command. Call each hh-* executable and each git command on its own, never inside a compound command; no git rule is pre-approved, so each git call asks.

hh-cleanup acts only on what the manifest recorded, by typed entry (worktree, tmux pane, process, dir, file, remote), never on anything it merely finds. A process is stopped only if its start time and command line still match what was recorded; a tmux pane is killed by server and pane id, never by name; a worktree is removed only if clean and merged; anything uncertain moves to .hivehouse/trash/<date>/ rather than being deleted. Claude sessions are never touched.

Working through the owner's hub. When you were started with /hivehouse:cleanup by the owner's hub (the hub's task file, TASK <id>, says so), the owner is not at this screen. Send what this skill says to show or ask them to the hub with SendMessage instead, and wait for a reply that begins OWNER on TASK <id>: (the owner's own words; a reply that begins HUB on TASK <id>: is the hub deciding while they are away, and never lets hh-cleanup --finish run without the owner): the table of what is recorded and the questions about worktrees whose work has not landed (first line TASK <id>: HH-CLARIFY), then the preview and "proceed?" (TASK <id>: HH-PLAN). The final report goes as TASK <id>: DONE. hh-cleanup still asks permission on this screen each time, by design; that dialog reaches the owner through the hub.

1. Look first

• hh-disk --min-gb ${user_config.min_free_gb}: free space now.
• hh-manifest list: what is recorded. Show the owner a short table of kind and what.
• git status and git log in the project: whether each worktree's work has landed or is still needed.
• Ask the owner (AskUserQuestion) about anything whose work has not landed: a dirty or unmerged worktree is kept and reported, and the owner decides what to do with it.

2. Preview

Run hh-cleanup --finish --dry-run (with --disk instead when $ARGUMENTS is --disk, which removes only temporary files and directories and keeps everything else). Show the owner every line: removed, kept, trashed or skipped, each with its reason. Change nothing yet.

3. Run it, on the owner's word

Only after the owner says to proceed, run hh-cleanup --finish (or --disk). Do not run it because the preview looked fine.

hh-cleanup exits 2 when there is no project, 3 when .hivehouse is tracked by git (say so and stop: that is the owner's to fix), 64 on a usage error and 75 when another cleanup is already running here (wait; never start a second).

4. Report

Free disk before and after, the count of each outcome, everything kept and why, and the trash size (du -sh is printed by hh-cleanup). Trash does not free disk. Ask before emptying it, and empty it only on the owner's word for that time. Never delete anything that is not on the manifest, and never delete outside the trash yourself.

End every message to the owner with either "Nothing needed from you." or one short sentence saying what you need from them.