fix(workstation): carry OS/sudo authz policy into managed-settings source + multi-tenancy doc

ADR-0015's policy change was applied live to /etc/claude-code/managed-settings.json, but that file self-deploys from the repo source scripts/workstation/managed-settings.json via the hourly reconcile (sync_managed_config). Without updating the source the next reconcile would REVERT /etc to the old 'never read other homes' rule. This updates the source-of-truth claudeMd (now byte-identical to /etc) so the change is durable + canonical, and refresh_codex_mirror propagates it to every user's ~/.codex/AGENTS.md. Also notes the access-model change in the multi-tenancy architecture doc (pointer to ADR-0015).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

docs/architecture/multi-tenancy.md

@@ -541,7 +541,7 @@ Separate from the in-cluster namespace-owner model above, the **devvm** (`10.0.1
 
 **RBAC tiers:** `admin` (Viktor — cluster-admin, unlocked tree, secrets) · `power-user` (cluster-wide read-only, NO Secrets, via a dedicated `oidc-power-user-readonly` ClusterRole) · `namespace-owner` (admin in own namespace only). Each session acts as the user's **own** OIDC identity (kubelogin), never the admin's.
 
-**Config inheritance (live):** wizard authors the base (his chezmoi-versioned `~/.claude`). Two native layers carry it to every user — the enforced org `claudeMd` in `/etc/claude-code/managed-settings.json` (top precedence, all sessions) and per-user `~/.claude/{skills,rules,…}` **symlinks** to the base (seeded via `/etc/skel`; edits propagate live). Secrets stay per-user at mode 600, never symlinked. **The managed config self-deploys from the repo** (2026-06-10): the hourly reconcile's `sync_managed_config` installs `scripts/workstation/managed-settings.json` to `/etc/claude-code/` whenever the repo copy changes — so editing the claudeMd = edit + commit, no manual install — and `refresh_codex_mirror` regenerates each user's `~/.codex/AGENTS.md` (a static mirror of the claudeMd; only files carrying the mirror header are touched, user-customized ones are left alone). Repo-level guidance (`.claude/CLAUDE.md`, `AGENTS.md`, `CONTEXT.md` in the infra repo) reaches non-admins through their auto-freshened clones — commit + push and every user has it within the hour.
+**Config inheritance (live):** wizard authors the base (his chezmoi-versioned `~/.claude`). Two native layers carry it to every user — the enforced org `claudeMd` in `/etc/claude-code/managed-settings.json` (top precedence, all sessions) and per-user `~/.claude/{skills,rules,…}` **symlinks** to the base (seeded via `/etc/skel`; edits propagate live). Secrets stay per-user at mode 600, never symlinked. **(2026-06-26: the managed `claudeMd` now defers OS-level file access to the OS/sudo — a user holding broad `sudo` may read other users' files incl. `~/.claude`; the mode-600 / no-symlink posture is unchanged but is no longer reinforced by an agent "never read other homes" rule. See [ADR-0015](../adr/0015-os-is-the-authorization-boundary.md).)** **The managed config self-deploys from the repo** (2026-06-10): the hourly reconcile's `sync_managed_config` installs `scripts/workstation/managed-settings.json` to `/etc/claude-code/` whenever the repo copy changes — so editing the claudeMd = edit + commit, no manual install — and `refresh_codex_mirror` regenerates each user's `~/.codex/AGENTS.md` (a static mirror of the claudeMd; only files carrying the mirror header are touched, user-customized ones are left alone). Repo-level guidance (`.claude/CLAUDE.md`, `AGENTS.md`, `CONTEXT.md` in the infra repo) reaches non-admins through their auto-freshened clones — commit + push and every user has it within the hour.
 
 **Memory — homelab CLI hooks (rolled out 2026-06-21, deploy-fixed 2026-06-22):** the per-user `claude_memory` MCP was retired for the **homelab-memory hooks** — the reconcile's `install_memory` (re)installs four scripts into `~/.claude/hooks/` each run (`homelab-memory-recall.py` UserPromptSubmit recall, `auto-learn.py` Stop-hook extraction, `pre-compact-backup.sh`/`post-compact-recovery.sh`), wires them into `settings.json` if-absent + additive, and removes the old `claude_memory` MCP. **The provisioner binary itself now self-deploys from the repo** (step 0: `bash -n`-gated `install` + re-exec when `scripts/t3-provision-users.sh` differs from `/usr/local/bin/t3-provision-users`, guarded against re-exec loops / DRY_RUN mutation) — added after this very rollout sat committed-but-undeployed for a day (only the manual `setup-devvm.sh` had ever deployed the binary), so the hourly reconcile kept running the pre-memory version and emo/anca silently lost memory (recall + auto-learn never wired). A latent `set -e` abort in `install_memory` (a bare `[[ -d plugin-dir ]] && …` returning non-zero) was also fixed; it had killed the reconcile after the first user the first time it actually ran. The hooks need a `MEMORY_API_KEY` (or `CLAUDE_MEMORY_API_KEY`) in the user's `settings.json` env — the `homelab` CLI defaults the API URL, so **the key is the only hard requirement**; `install_memory` reuses an existing key and only WARNs if absent (it does NOT mint one — that's an admin Vault step, see Remaining). wizard + emo carry a key from their original MCP setup; **ancamilea is keyless → her memory no-ops until a key is minted.** (`auto-learn.py`'s passive store calls the API directly, so it additionally needs `*_API_URL` in env to avoid its local-SQLite fallback; recall + manual `homelab memory store` go through the URL-defaulting CLI and need only the key.)
 

scripts/workstation/managed-settings.json

@@ -1,4 +1,4 @@
 {
-  "claudeMd": "# Viktor Barzin homelab — shared multi-user Claude Code Workstation (devvm)\n\nYou are running as a specific OS user on a SHARED devvm Workstation, not as the admin. These org-wide rules apply to EVERY user and sit at the top of settings precedence (they cannot be overridden by a user's own config):\n\n- Respect your permission tier. kubectl, Vault, and infra access are scoped to your RBAC tier (admin / power-user / namespace-owner). Do not attempt to escalate privileges or reach another user's resources.\n- Secrets are per-user. Never read another user's home directory, credentials, tokens, or ~/.claude secrets. Your own secrets live in your home at mode 600.\n- Infrastructure changes go through Terraform/Terragrunt — never direct kubectl apply/edit/patch. Committed stack changes are auto-applied by CI on push to master; verify the live result with your read-only kubectl.\n- The AGENT does ALL git mechanics silently — the user may not know git, so never ask them to commit, push, pull, or open anything, and never surface git jargon. Lifecycle (worktrees, landing, cleanup): ~/.claude/rules/execution.md. Org red-lines on top:\n  - THE COMMIT MESSAGE IS THE AUDIT TRAIL — subject says WHAT changed; body says WHY in plain words (paraphrase the user's actual request).\n  - Never use [ci skip] as a non-admin (it hides the change from the audit feed).\n  - Push rejected by branch protection (user not whitelisted) → fall back to a <os-user>/<topic> branch + PR via the Forgejo API (token = password field in ~/.git-credentials).\n  - Keep every clone on a clean master when done; tell the user in plain words what happened.\n  - Full recipe: AGENTS.md → \"Non-admin workstation users\" in your infra clone.\n- Follow the engineering rules in ~/.claude/rules/ (execution, planning) and every CLAUDE.md in the repo tree.\n- Code lives under ~/code in one of two per-user layouts: either ~/code IS the git-crypt-LOCKED infra clone (single layout), or ~/code is a workspace directory of per-project clones — the locked infra clone at ~/code/infra plus other project repos alongside it. [ -d ~/code/.git ] means single. In locked infra clones secret files read as ciphertext — that is expected, not an error.\n",
+  "claudeMd": "# Viktor Barzin homelab — shared multi-user Claude Code Workstation (devvm)\n\nYou are running as a specific OS user on a SHARED devvm Workstation, not as the admin. These org-wide rules apply to EVERY user and sit at the top of settings precedence (they cannot be overridden by a user's own config):\n\n- Respect your permission tier for cluster/infra. kubectl, Vault, and infra access are scoped to your RBAC tier (admin / power-user / namespace-owner) — don't try to exceed it. At the OS level, authorization follows the OS: where your user holds `sudo`, using it is permitted and audited.\n- OS access follows OS permissions, not a separate rule here. You may read whatever your OS user can read — directly or via `sudo` where you hold it — including other users' home directories, credentials, tokens, and `~/.claude`. Do not impose restrictions stricter than the OS; no extra prompt is required for OS-authorized access. Protect your own secrets at mode 600.\n- Infrastructure changes go through Terraform/Terragrunt — never direct kubectl apply/edit/patch. Committed stack changes are auto-applied by CI on push to master; verify the live result with your read-only kubectl.\n- The AGENT does ALL git mechanics silently — the user may not know git, so never ask them to commit, push, pull, or open anything, and never surface git jargon. Lifecycle (worktrees, landing, cleanup): ~/.claude/rules/execution.md. Org red-lines on top:\n  - THE COMMIT MESSAGE IS THE AUDIT TRAIL — subject says WHAT changed; body says WHY in plain words (paraphrase the user's actual request).\n  - Never use [ci skip] as a non-admin (it hides the change from the audit feed).\n  - Push rejected by branch protection (user not whitelisted) → fall back to a <os-user>/<topic> branch + PR via the Forgejo API (token = password field in ~/.git-credentials).\n  - Keep every clone on a clean master when done; tell the user in plain words what happened.\n  - Full recipe: AGENTS.md → \"Non-admin workstation users\" in your infra clone.\n- Follow the engineering rules in ~/.claude/rules/ (execution, planning) and every CLAUDE.md in the repo tree.\n- Code lives under ~/code in one of two per-user layouts: either ~/code IS the git-crypt-LOCKED infra clone (single layout), or ~/code is a workspace directory of per-project clones — the locked infra clone at ~/code/infra plus other project repos alongside it. [ -d ~/code/.git ] means single. In locked infra clones secret files read as ciphertext — that is expected, not an error.\n",
   "model": "claude-opus-4-8"
 }