docs(architecture): Etap 0 changes appendix in recon; telegram-frontend pattern in legacy doc

RECON appendix summarizes the whole etap-0 cleanup (topology dormant status,
hosts reconciliation, dead-code removal, legacy decision), carries the
runbook line for the stale chelsty pending action
(rm /opt/homelab/actions/pending/redeploy-chelsty-infra-ha-diag-agent.json —
verified the only chelsty-targeted pending file 2026-07-30), and lists
discrepancies found but out of scope: lustro watchtower, stale solaria/chelsty
world-state keys, piha naming-drift duplicate keys, gokapi still down with a
stuck pending redeploy, outline/joplin/umami still lacking services/ dirs.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

docs/architecture/RECON-multiagent-2026-07-27.md

@@ -549,3 +549,91 @@ Decisions a human must make; no recommendations attached.
    accepted policy and which are defects, and whether "event emitted but
    unroutable" should be visible anywhere. At stake: the difference between
    "monitored" and "logged".
+
+---
+
+## Etap 0 changes (2026-07-28)
+
+Repo-only truth cleanup driven by this recon (commits 2026-07-28 → 2026-07-30,
+branch `task/porzadki-topologia` plus earlier etap-0 commits already on master).
+No runtime state was touched.
+
+**Topology / nodes**
+
+- `inventory/topology.yaml`: chelsty-infra and chelsty-ha marked
+  `status: dormant` (site hardware down since ~2026-06-01, pending physical
+  revival — F20.12 / open question 6); per-node service enumerations removed —
+  `hosts/<node>/services.yaml` is authoritative (F20.1-2); `deployment.mode`
+  corrected to `push` (F20.10); lustro added as a full monitored node with its
+  duty cycle documented (nightly ~23:30 power-off, one liveness cycle per day —
+  same pattern as solaria, F20.11).
+- Supervisor reads `status: dormant` from `inventory/topology.yaml`
+  (`_load_dormant_nodes`) and skips desired-state loading, disk-cleanup and
+  node-alert generation for dormant nodes — replacing the hardcoded
+  chelsty-name checks (topology is readable from the supervisor's runtime
+  context, so no env-var fallback was needed).
+
+**hosts/ reconciliation (B6/F20)**
+
+- `hosts/solaria/services.yaml`: added stability-agent, node_exporter, and
+  planner-agent (`monitor: false`, legacy ai-cluster family, retirement
+  candidate).
+- `hosts/vps/services.yaml`: added stability-agent, npm, outline,
+  joplin-server, umami (all verified healthy in world state 2026-07-30);
+  humanai-mailer / humanai-landing added as `unmanaged: true` +
+  `monitor: false` (hand-run, no compose labels); ai-cluster deliberately gets
+  NO entry — comment block points at `ai-cluster-LEGACY.md`.
+- `hosts/piha/services.yaml`: single comment block enumerating the ~28 known
+  unmanaged containers (B5) — bringing them in is a later stage (open
+  question 5).
+- `hosts/saturn/services.yaml`: created with an explicit empty service list
+  (dev workstation / orchestrator, nothing monitored — F20.4).
+- `hosts/lustro/services.yaml`: added node-exporter and piper-tts (verified
+  running); watchtower noted as deliberately unmanaged.
+- `services/stability-agent/service.yaml`: `owner_node: chelsty` →
+  `per-host` (B7/F20.5).
+- `services/mosquitto/service.yaml`: marked NOT DEPLOYED / legacy manifest —
+  matches nothing that runs (C8/F20.6); kept for reference pending open
+  question 7.
+
+**Dead code**
+
+- `scripts/deploy/deploy-role.sh` deleted (referenced nonexistent `roles/`,
+  B7).
+- Supervisor: `mqtt_unreachable` removed from `CONTAINER_RESTART_TRIGGERS`
+  (dead branch — the observer never creates incidents with that trigger_type,
+  D15).
+
+**Legacy**
+
+- `docs/architecture/ai-cluster-LEGACY.md`: ai-cluster is retired in place,
+  not migrated (bus idle since 2026-06-09, C9); branch `task/ai-cluster-solaria`
+  stays unmerged as documentation; surviving patterns listed; runtime
+  retirement runbook (stop stack on vps, observe `free -m`, remove containers)
+  to be executed in a separate supervised session.
+
+**Runbook — stale chelsty action (from 1c, do in a runtime session)**
+
+- On vps: `rm /opt/homelab/actions/pending/redeploy-chelsty-infra-ha-diag-agent.json`
+  — the only chelsty-targeted pending action as of 2026-07-30 (verified via
+  `grep -l chelsty /opt/homelab/actions/pending/*.json`).
+
+**Discrepancies found during this pass, NOT covered by etap 0**
+
+- lustro also runs `watchtower` (auto-updater) — undocumented anywhere;
+  whether unattended container updates on a monitored node are policy needs a
+  decision.
+- World state carries stale solaria keys (`solaria/executor`,
+  `solaria/operator-ui`, `solaria/kb-postgres`, `solaria/paperless-worker`) for
+  containers that do not run there (B5 counted 5) — world state never prunes
+  departed services; same mechanism keeps all `chelsty-infra/*` keys alive.
+- piha world state holds duplicate keys from naming drift
+  (`piha/node-exporter` vs `piha/node_exporter`, `immich-server` vs
+  `immich_server`, generic `app`/`db`/`database`/`broker` keys) — event-source
+  naming is not normalized.
+- `gokapi` remains desired-but-down with its `redeploy` action stuck in
+  `pending` (F20.3); the wider undrained queue (18 pending actions) is open
+  question 4 — untouched here.
+- `outline`, `joplin`, `umami` run on vps with no `services/<name>` dir on
+  master — their hosts entries added in etap 0 document this; actual GitOps
+  cutover (compose in repo) is still pending (F20.2).

docs/architecture/ai-cluster-LEGACY.md

@@ -32,7 +32,9 @@ z ai-clustra **wzorce projektowe**:
 - allowlisty kształtu komend przed wykonaniem czegokolwiek na hoście — z lekcją
   z reconu (A2): allowlist musi być **egzekwowana i celowana** (martwa stała
   `SERVICE_NAMES` + substring-match po `docker ps` to antywzorzec),
-- tryb preview/diagnose przed wykonaniem.
+- tryb preview/diagnose przed wykonaniem,
+- frontend telegramowy jako kanał operatorski (wzorzec żyje dalej w
+  agent-system/telegram-bot na piha — ten zostaje).
 
 Kod ai-clustra **nie jest** przenoszony.