Phase 4 infrastructure-as-code for the nextcloud-todos service (watches the
Nextcloud Personal task list; classifies todos via local qwen3-8b and routes
research/mutating work through claude-agent-service). Clones the
recruiter-responder service pattern end-to-end. Written only — NOT applied.
- stacks/nextcloud-todos/{main.tf,terragrunt.hcl}: new aux stack cloning
recruiter-responder — ns (tier aux, istio-injection disabled, keel enrolled),
two ExternalSecrets (vault-kv app secrets + vault-database DSN), Recreate
deployment with alembic-migrate init-container, ClusterIP svc, /cb-only
HMAC-gated ingress (auth=none, proxied), and an idempotent webhook-register
null_resource (OCS webhook_listeners API, both CalendarObject Created/Updated
events -> internal svc URL, Bearer auth).
- stacks/vault/main.tf: pg_nextcloud_todos static role (nextcloud_todos, 7d
rotation) + pg-nextcloud-todos in the postgresql allowed_roles array.
- stacks/dbaas/modules/dbaas/main.tf: pg_nextcloud_todos_db null_resource
(clone of pg_tripit_db) — creates role+DB, pins role search_path, and
creates schema nextcloud_todos AUTHORIZATION nextcloud_todos.
- stacks/openclaw/main.tf: install-nextcloud-todos-plugin init-container,
nextcloud-todos-api in plugins.allow + the doctor-fix re-add + plugins
enable, NEXTCLOUD_TODOS_URL/NEXTCLOUD_TODOS_TOKEN env, and the cross-path
ESO key (secret/nextcloud-todos.webhook_bearer_token).
- stacks/uptime-kuma/modules/uptime-kuma/main.tf: internal /healthz HTTP
monitor. Prometheus /metrics scrape via svc annotations in the new stack.
- .gitleaksignore: allowlist two curl-auth-user false positives (the OCS
webhook curl uses a Vault-sourced shell var, not a literal credential).
KV seed (secret/nextcloud-todos) + applies are deferred to the apply runbook.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
c958f6a589