murmurent-reset

name	murmurent-reset
description	Back up, then reset murmurent machine state to a fresh start so `murmurent centre-init` runs first-run again. Tiered (centre / install / full) with a mandatory backup, a dry-run preview, and credentials + other-project installs protected behind explicit --nuke flags. Use when the user wants a clean slate / fresh copy from the repo.
user_invocable	true

Reset this machine's murmurent state to a clean starting point — safely. The default is the least destructive thing (centre), a backup is always taken first, and the genuinely dangerous deletions (credentials, other projects' installs) never happen without an explicit flag.

The heavy lifting is in reset.sh (next to this file). Your job is to pick the level from what the user asked, preview with --dry-run, get confirmation, then run for real.

Levels

Level	Removes	Keeps
centre (default)	~/.murmurent/lab_info/ only → centre-init is first-run again	everything else
install	centre + reinstall the tool from ~/repos/murmurent (uv tool install --force --python 3.12 -e '.[dashboard,slack,mcp]') + scripts/setup.sh + murmurent install --hooks	credentials, installations, audit
full	install + machine-local caches (workspaces/, *.log, dashboard.pid, security/agent_cache, stale RESUME.md)	credentials, installations/, decommissions/, audit logs, hosts.yaml/machine.yaml
data	all data you entered into ~/.murmurent — lab_info/, profile.yaml, hosts/machine yaml, inventory/, cores/, onboarding/, decommissions/, security/, identity/cards/trust/revocation, logs — everything except key material (allowlist-based, so new files are caught)	keys/, age/, installations/ (other projects), and ~/.config/murmurent. No reinstall.

Use data for "wipe everything I've entered and start over, but keep my keys and credentials." It's the level to reach for when full leaves too much behind (full keeps profile.yaml, hosts.yaml, inventory/, etc.).

Opt-in extras (only with the flag): --nuke-installations (also wipes ~/.murmurent/installations/ — other projects' manifests), --nuke-credentials (also wipes ~/.config/murmurent/ — the slack-tokens + keys), --nuke-keys (with --level data, also removes ~/.murmurent/keys/

• age/ for a fully fresh identity; default keeps them), --nuke-labs (also removes this machine's lab-management repos ~/repos/murmurent_lab_mgmt_* (legacy: ~/repos/wigamig_*) — they hold the roster; see below), --uninstall (first completely removes the existing murmurent install — the uv-tool one and stray conda/pipx copies that shadow it — before any reinstall).

Lab repos (~/repos/murmurent_lab_mgmt_* (legacy: ~/repos/wigamig_*)) — the roster

A PI's lab-management repo (~/repos/murmurent_lab_mgmt_<group> (legacy ~/repos/wigamig_<group>), created by pi-init / murmurent init) holds members/*.md — the roster, which is the source of truth for member identity (card fingerprints included). It lives under ~/repos, which reset never touches by default, so every reset just lists these repos and leaves them alone.

--nuke-labs removes them, but safely: each is tar'd into the backup dir first, and any repo with uncommitted or unpushed work is refused (you'll be told to push/commit it first). A standalone lab with no git remote is always refused — its roster is the only copy. After a successful nuke the dangling lab_mgmt_path pointer is dropped, so pi-init is first-run again. Only reach for --nuke-labs when the user explicitly wants the lab's roster gone from this machine and has confirmed it's pushed somewhere safe.

Completely removing the old install

When the user asks to "completely remove" / "uninstall" murmurent:

• Remove it and leave nothing: --level centre --uninstall. This uninstalls every murmurent executable and does not reinstall — the machine ends up with no murmurent on PATH.
• Remove the old one and put back a clean one (the usual "fresh copy from the repo" intent): --level install --uninstall — uninstalls first, then reinstalls editable from ~/repos/murmurent.

--uninstall is what handles the stray-duplicate-install problem (e.g. an old pip install -e copy in a conda env shadowing the uv-tool one). It scans conda base/envs + pipx and removes any murmurent it finds. Working clones under ~/repos/* are never touched — only installed executables.

How to run it

1. Pick the level. No level / "reset the centre" / "start centre-init fresh" → centre. "reinstall / fresh copy from the repo" → install. "full machine reset / everything clean" → full. If the user's words imply removing credentials or other projects, ask before adding a --nuke flag — never infer those.

2. Dry-run first, always. Run: bash ~/.claude/skills/murmurent-reset/reset.sh --level <lvl> --dry-run [nukes] (or the repo path skills/murmurent-reset/reset.sh). Show the user the exact list of what it would remove and the backup path.

3. Confirm, then execute. Only after the user is happy, re-run with --yes instead of --dry-run. The script refuses to do anything destructive without --yes, so this two-step is enforced, not just convention.

4. Report. Show the backup tarball path (~/.murmurent_backups/reset_*.tgz) and the restore one-liner, then confirm murmurent centre-status reports "no centre initialised". If the level was install/full, confirm murmurent dashboard --hifi still launches.

Guardrails (the script enforces these — don't work around them)

• Backup is mandatory and lands OUTSIDE ~/.murmurent (~/.murmurent_backups/) so even a full wipe can't delete it. Never pass a flag to skip it; there isn't one.
• Never touches ~/repos/* working clones, ~/.claude/CLAUDE.md, ~/.claude/memory/, or ~/.claude/projects/ — with one opt-in exception: --nuke-labs removes ~/repos/murmurent_lab_mgmt_* (legacy: ~/repos/wigamig_*) lab-management repos (backed up first, and only when they have no uncommitted/unpushed work). Nothing else under ~/repos is ever touched.
• Credentials and other-project installs are preserved by default. On this machine ~/.config/murmurent/ slack-tokens + keys/ are real, and ~/.murmurent/installations/ holds manifests for the user's other projects (manuscript, mp3/mp4, …). Removing either needs the explicit --nuke-* flag AND a clear go-ahead from the user.
• If the user says "wipe everything / pristine machine", read that back as "this includes your slack-token, keys, and every other project's install — confirm?" before adding both --nuke flags.