Reconciliation routine¶

murmurent reconcile compares Murmurent's recorded state to on-disk reality across every registered host and reports drift. Default is dry-run; --apply repairs the actionable subset.

Before the detectors run, reconcile fast-forwards the local lab_mgmt clone (git pull --ff-only) so the roster and cert-project registry reflect what the PI last pushed: this is what keeps a member's Lab Members panel current without clicking its update button. Offline, diverged, or not-a-git-clone lab_mgmt is a note in the report, never a failure.

What counts as drift¶

Kind	Severity	Auto-repair?	Repair action
`orphan_installation`	actionable	✓	Move `~/.murmurent/installations/<name>.yaml` into `installations/.archive/<name>_<date>.yaml`
`orphan_registry`	actionable	✓	Set `status: archived` + `archived_at: <date>` in the lab_mgmt registry frontmatter (file preserved, lab history is shared)
`missing_charter`	warn	✗	User decides: re-adopt the clone, or remove from Murmurent
`unadopted_clone`	info	✗	Click ↑ adopt in the Repos panel
`lab_mgmt_uncommitted`	warn	✗	Review + commit + push lab_mgmt: local-only edits are invisible to the lab (roster writers auto-commit, so this usually means a hand-edit)
`lab_mgmt_unpushed`	warn	✗	`git -C <lab-mgmt> push`: commits exist locally that members can't pull yet

Remote (SSH) hosts are probed in a single batched bash call per host. A transient SSH failure (host unreachable) is conservative: no installations on that host are reported as orphaned. That way lab-server being down for a reboot won't auto-deactivate everything.

Manual run¶

murmurent reconcile                  # dry-run; exit 1 if actionable drift
murmurent reconcile --apply          # repair actionable findings
murmurent reconcile --slack-body     # also print a Slack-formatted summary

Exit codes: - 0: clean OR everything actionable was applied. - 1: actionable drift exists and --apply wasn't passed (so cron / CI can branch on it).

Members' profile updates (PI only)¶

On the PI's machine, reconcile also lists members' profile changes waiting as issues on the lab management repo, and murmurent reconcile --apply applies them to the roster. See How a member's profile edit reaches the roster.

Daily schedule via CC `/routine`¶

Set up once:

/routine create murmurent-reconcile
  prompt: |
    Run `murmurent reconcile --slack-body` in the shell and capture the
    output. If exit code is 1 (actionable drift was found in dry-run
    mode), post the slack-body block to #claude-test via
    mcp__claude_ai_Slack__slack_send_message and STOP — do not
    --apply automatically. If exit code is 0 and the report shows
    info-only findings (unadopted clones), post a short summary and
    stop. Never run --apply without my confirmation.
  schedule: daily at 09:00 America/Toronto

Why dry-run on the cron and --apply is manual: a false positive that auto-archives an active installation manifest is disruptive and easily missed. The routine surfaces drift in Slack; you decide whether to run murmurent reconcile --apply from a terminal.

Recovery from auto-archive¶

Archived manifests live in ~/.murmurent/installations/.archive/ with a date suffix. To restore: copy the file back to ~/.murmurent/installations/<name>.yaml.

Archived registry entries keep status: archived in the frontmatter. To restore: remove the status: and archived_at: lines from the lab_mgmt registry .md file. Commit + push the lab_mgmt repo so other group members see the restoration.