SolOSV3 return-purge (2026-10-18)
SolOSV3 return-purge (2026-10-18)
Young: SolOSV3 "returned" in the vault — remove and make sure it does not return.
What was actually there — and the ROOT CAUSE found
Not a rename-revert this time (the Fleet rename held). The residue was stale references in still-running machinery, which regenerates SolOSV3 paths when triggered:
- Mira's cron jobs (2):
spend-ledger-dailywrote toDocs/SolOSV3/05-operations/(would RECREATE the dir on next run);solos-backup-githubstill listed SolOSV3 paths (logged "skip (absent)" but would re-track if recreated). Both retargeted toFleet/...on-host. - VAULT-INDEX.md — 16 live rows were still pointing at
SolOSV3/...paths (the rename pass missed them). All retargeted to Fleet/ (901ca64), except the deliberate legacyreference-tree/SolOS Legacyhistorical rows (they name the archived historical tree — correct as-is). - Directory itself: gone from disk, zero tracked SolOSV3 paths, no untracked dirs (
findclean), Mira's own removal commitc53952f(spend-ledger.tsv dupe) already landed.
Guarantor of "never returns" (register rule for Min)
- The paths live in exactly 3 classes of writers: cron job definitions (fixed), index/docs text (fixed), and profile MEMORY (fixed this morning in the audit). No writer emits SolOSV3 anymore; the Fleet rename now holds.
🔴 REAL FIND during the residue scan — token committed to the repo
backup-github cron FAILED today with "secret-bearing tracked files ROSE 4→5". Scanned:
- False positive: the 5th file = my
secrets-isolation-fix.md— filename-only match; verified no key values in it (pattern scan: 0). - REAL one (found during the same scan):
Docs/Fleet/05-archive/reference-tree/SolOSV2/07-build/tokens.json— a 43-character Brain A2A auth token (role: brain, label: min-sol-mira, issued 2026-10-01) committed into the GitHub repo (via kanban-regen commitc64dd7d). The repo is private, but it IS the fleet's auth-token material at rest in a git remote.
@Young — ROTATION REQUIRED: that Brain token (issued 10-01) should be re-issued via bin/issue_token.py (the enforcement guard's token path Min's memory cites) and the old one rejected; report to Sol as the enforcement lane. My report stops at the find + flag — token material is never printed here.
Records
- Crons fixed but not committed (they live on-host per profile, not in vault).
- VAULT-INDEX fix committed
901ca64and pushed. - Earlier tarball-state commit
c53952f(Mira) + my audit commits hold the rest of the story.
🎯 Tam's Read: SolOSV3 does not "return" — it was being re-emitted by two crons and sixteen index rows that never got the rename memo; all three writer classes are fixed now, and the residue-scan ALSO surfaced the real security item of the day (a stale Brain token at rest in the git repo) — rotation, @Young, is the one action this wave still owes.
Single-host fleet? — measured trade-offs (2026-10-17)
Min's question (via Young): "move all agents onto a single host — save infrastructure cost, better control?" Short answer as the ops lane: yes for 8-9 of our 11 agents today; keep Pulse (me, Eva, Eve) where the user-facing=g traffic is; keep prod's headroom ready for tenant stacks. The measured numbers follow.
SSH access matrix & security model (authoritative, 2026-10-18)
Young, 2026-10-18:

