Harden Cloud Agent env secrets and email_records.is_read migration

.cursor/environment.json

@@ -0,0 +1,20 @@
+{
+  "name": "Naruon dev (backend + frontend + Postgres/pgvector)",
+  "user": "ubuntu",
+  "install": "bash .cursor/install.sh",
+  "start": "bash .cursor/start.sh",
+  "terminals": [
+    {
+      "name": "backend",
+      "command": "cd backend && . .venv/bin/activate && python scripts/start_backend.py --host 0.0.0.0 --port 8000"
+    },
+    {
+      "name": "frontend",
+      "command": "cd frontend && corepack pnpm@11.5.3 dev --port 3000 --hostname 0.0.0.0"
+    }
+  ],
+  "ports": [
+    { "name": "backend", "port": 8000 },
+    { "name": "frontend", "port": 3000 }
+  ]
+}

.cursor/install.sh

@@ -0,0 +1,43 @@
+#!/usr/bin/env bash
+# Idempotent repository bootstrap for Naruon Cloud Agent environments.
+#
+# Prepares durable, source-derived state after checkout:
+#   * system packages (PostgreSQL 16 + pgvector, Python venv/build tooling)
+#   * the backend virtualenv + pinned Python requirements
+#   * the frontend pnpm dependency tree
+#
+# Per-boot service startup (Postgres, schema migrations, dev secrets) lives in
+# .cursor/start.sh so it re-runs on every VM boot, including builds.
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+cd "$REPO_ROOT"
+
+echo "==> [install] system packages (postgresql-16, pgvector, python venv, build tools)"
+export DEBIAN_FRONTEND=noninteractive
+sudo apt-get update -qq
+sudo apt-get install -y -qq \
+  postgresql-16 \
+  postgresql-16-pgvector \
+  postgresql-client-16 \
+  python3.12-venv \
+  python3.12-dev \
+  build-essential
+
+echo "==> [install] backend virtualenv + requirements"
+cd "$REPO_ROOT/backend"
+if [ ! -x ".venv/bin/python" ]; then
+  python3 -m venv .venv
+fi
+# shellcheck disable=SC1091
+. .venv/bin/activate
+python -m pip install --upgrade pip
+python -m pip install --require-hashes -r requirements-hashes.txt
+
+echo "==> [install] frontend dependencies (pnpm@11.5.3)"
+cd "$REPO_ROOT/frontend"
+corepack enable
+corepack prepare pnpm@11.5.3 --activate
+corepack pnpm@11.5.3 install --frozen-lockfile
+
+echo "==> [install] done"

.cursor/start.sh

@@ -0,0 +1,80 @@
+#!/usr/bin/env bash
+# Per-boot runtime reconciliation for Naruon Cloud Agent environments.
+#
+# Runs on every VM start (idempotent): brings up the PostgreSQL cluster,
+# materializes a local dev .env with generated secrets on first boot, ensures
+# the app database + pgvector extension exist, and applies Alembic migrations.
+#
+# Dependency installation lives in .cursor/install.sh; this script only
+# reconciles per-boot state and then returns so the backend/frontend terminals
+# can start.
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+cd "$REPO_ROOT"
+
+ENV_FILE="$HOME/.env"
+PY="$REPO_ROOT/backend/.venv/bin/python"
+
+echo "==> [start] ensuring PostgreSQL 16 cluster is online"
+if ! pg_lsclusters -h 2>/dev/null | awk '{print $4}' | grep -q online; then
+  sudo pg_ctlcluster 16 main start || true
+fi
+# Wait for the server to accept connections before touching it.
+for _ in $(seq 1 30); do
+  if sudo -u postgres pg_isready -q; then break; fi
+  sleep 1
+done
+if ! sudo -u postgres pg_isready -q; then
+  echo "==> [start] PostgreSQL did not become ready" >&2
+  exit 1
+fi
+
+echo "==> [start] generating local dev .env on first boot (secrets are per-VM)"
+if [ ! -f "$ENV_FILE" ]; then
+  "$PY" - "$ENV_FILE" <<'PYGEN'
+import os, secrets, sys
+from pathlib import Path
+from cryptography.fernet import Fernet
+
+env_path = Path(sys.argv[1])
+db_password = secrets.token_urlsafe(24)
+hmac_secret = secrets.token_urlsafe(48)
+enc_key = Fernet.generate_key().decode()
+
+env_path.write_text(
+    "# Naruon local dev environment (generated per-VM; not committed).\n"
+    f"DATABASE_URL=postgresql+asyncpg://postgres:{db_password}@127.0.0.1:5432/ai_email\n"
+    f"AUTH_SESSION_HMAC_SECRET={hmac_secret}\n"
+    f"ENCRYPTION_KEY={enc_key}\n"
+    "DEBUG=false\n"
+    "RUNTIME_ENVIRONMENT=development\n"
+    "ENABLE_PROMETHEUS_METRICS=false\n"
+    "ALLOWED_CORS_ORIGINS=http://localhost:3000,http://127.0.0.1:3000,"
+    "http://localhost:8000,http://127.0.0.1:8000\n"
+    "SMTP_MODE=simulated\n"
+    "OPENAI_API_KEY=\n"
+    "OPENAI_EMBEDDING_MODEL=text-embedding-3-small\n"
+    "OPENAI_MODEL=gpt-4o\n",
+    encoding="utf-8",
+)
+os.chmod(env_path, 0o600)
+print(f"wrote {env_path}")
+PYGEN
+fi
+
+echo "==> [start] reconciling database role, database, and pgvector extension"
+# Keep the local postgres role secret in sync with DATABASE_URL without
+# interpolating it into SQL or the process argument list.
+"$PY" "$REPO_ROOT/backend/scripts/reconcile_local_postgres_role.py" --env-file "$ENV_FILE"
+if ! sudo -u postgres psql -tAc "SELECT 1 FROM pg_database WHERE datname='ai_email'" | grep -q 1; then
+  sudo -u postgres createdb ai_email
+fi
+sudo -u postgres psql -d ai_email -v ON_ERROR_STOP=1 \
+  -c "CREATE EXTENSION IF NOT EXISTS vector;" >/dev/null
+
+echo "==> [start] applying database migrations (alembic upgrade head)"
+cd "$REPO_ROOT/backend"
+"$PY" scripts/migrate_db.py
+
+echo "==> [start] done"

AGENTS.md

@@ -348,6 +348,12 @@ in this repo.
   embedding generation is unavailable. Tests must cover the local
   `embeddinggemma` path so Data workspace imports do not silently bypass the
   selected embedding model.
+- PostgreSQL `text` / `hashtext()` cannot encode a NUL (`0x00`) octet. Do not
+  pass a raw `user_id\\x00organization_id` (or any other NUL-separated payload)
+  as a `pg_advisory_lock` bind value. Derive a NUL-free digest such as
+  SHA-256 hex and keep mocked/SQLite tests from asserting the raw NUL form —
+  those dialects skip the lock and hide `CharacterNotInRepertoireError` until
+  a real Postgres import.
 - Home/Today dashboard reply-wait surfaces must read signed
   `/api/emails/pending-replies` data instead of inferring pending replies from
   generic inbox fixtures or static copy. Tests and E2E mocks must verify the

ARCHITECTURE.md

@@ -145,6 +145,12 @@ remains a local/dev compatibility path that creates the `vector` extension,
 metadata-defined tables for fresh local databases, and idempotent backfills for
 existing local databases.
 
+Cloud Agent VMs use the same managed migration path through
+`.cursor/start.sh` → `scripts/migrate_db.py`. Local role-secret alignment is
+stdin dollar-quoted SQL (`scripts/reconcile_local_postgres_role.py`); do not
+interpolate `DATABASE_URL` secrets into `psql -c`. See
+[`docs/development/cloud-agent-environment.md`](docs/development/cloud-agent-environment.md).
+
 ## Send boundary
 
 Outbound replies preserve `In-Reply-To` and `References` headers in the built

CHANGELOG.md

@@ -1,4 +1,5 @@
 ## [Unreleased]
+- Cloud Agent `.cursor/start.sh`는 `DATABASE_URL` 역할 비밀을 `psql -c`에 보간하지 않고 `scripts/reconcile_local_postgres_role.py`의 dollar-quoted stdin으로 맞춥니다. `install.sh`는 `requirements-hashes.txt`를 `--require-hashes`로 설치합니다. 신선한 DB에서 폐기된 `emails` 테이블을 가정하던 `0011_email_read_state`는 테이블이 없으면 no-op이고, `0018_email_record_read_state`가 정식 `email_records.is_read DEFAULT true`를 가드로 맞춥니다. 가져오기 quota advisory lock 키는 NUL 없는 SHA-256 hex입니다.
 - EmailDetail 테스트가 지원하지 않는 스레드 병합/분리 버튼을 `textContent`뿐 아니라 `aria-label`과 `title` 접근 가능 이름으로도 검출하도록 바꿔, 아이콘 전용 버튼 회귀를 놓치지 않습니다.
 ### 주제 측정 경계 (Topic Measurement)
 

CLAUDE.md

@@ -98,6 +98,17 @@ The other compose files are purpose-specific evaluation/evidence stacks:
 defaults anywhere; Compose/Kubernetes/operators must inject them, and startup
 fails closed when they are missing.
 
+### Cloud Agent environment (`.cursor/`)
+
+Repo-managed Cloud Agent bootstrap lives in `.cursor/environment.json`.
+`install.sh` installs PostgreSQL 16/pgvector and hashed Python deps;
+`start.sh` brings Postgres online, mints a per-VM `~/.env`, and runs
+`scripts/migrate_db.py`. Role-secret sync goes through
+`backend/scripts/reconcile_local_postgres_role.py` (dollar-quoted stdin, never
+`psql -c` interpolation). After boot, open http://127.0.0.1:3000 and sign in
+with a minted HMAC session — see
+`docs/development/cloud-agent-environment.md`.
+
 ## Architecture
 
 Naruon is an AI email workspace: a web client/control plane over

backend/alembic/versions/0011_email_read_state.py

@@ -14,6 +14,19 @@
 
 
 def upgrade() -> None:
+    # ``emails`` was the pre-reconciliation email table. ``email_records`` is now
+    # the single source of truth and already declares ``is_read`` via the model
+    # metadata created in 0001, so this legacy column add only applies to older
+    # databases that still carry the retired ``emails`` table. Guard on the table
+    # existing (matching the has_table/has_column pattern used by later
+    # revisions) so ``alembic upgrade head`` succeeds on fresh databases.
+    bind = op.get_bind()
+    inspector = sa.inspect(bind)
+    if not inspector.has_table("emails"):
+        return
+    existing_columns = {column["name"] for column in inspector.get_columns("emails")}
+    if "is_read" in existing_columns:
+        return
     op.add_column(
         "emails",
         sa.Column(
@@ -26,4 +39,7 @@ def upgrade() -> None:
 
 
 def downgrade() -> None:
-    op.drop_column("emails", "is_read")
+    # No-op: this revision only reconciles a retired ``emails`` table. Dropping
+    # ``is_read`` when the column is present would also remove a pre-existing
+    # column this revision did not create.
+    return

backend/alembic/versions/0018_email_record_read_state.py

@@ -0,0 +1,62 @@
+"""Guard email_records.is_read with NOT NULL DEFAULT true.
+
+Revision ID: 0018_email_record_read_state
+Revises: 0017_merge_newsdom_carddav_heads
+
+``0011_email_read_state`` only mutates the retired ``emails`` table. Fresh
+databases already receive ``email_records.is_read`` from ``0001``
+``create_all`` plus the current model ``server_default``. Existing databases
+whose ``email_records`` row predates that column (or lacks a server default)
+still need a guarded additive revision so raw INSERTs that omit ``is_read``
+succeed.
+"""
+
+from __future__ import annotations
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = "0018_email_record_read_state"
+down_revision = "0017_merge_newsdom_carddav_heads"
+branch_labels = None
+depends_on = None
+
+_EMAIL_RECORDS_TABLE = "email_records"
+_READ_STATE_COLUMN = "is_read"
+
+
+def upgrade() -> None:
+    bind = op.get_bind()
+    inspector = sa.inspect(bind)
+    if not inspector.has_table(_EMAIL_RECORDS_TABLE):
+        return
+    columns = {
+        column["name"]: column
+        for column in inspector.get_columns(_EMAIL_RECORDS_TABLE)
+    }
+    if _READ_STATE_COLUMN not in columns:
+        op.add_column(
+            _EMAIL_RECORDS_TABLE,
+            sa.Column(
+                _READ_STATE_COLUMN,
+                sa.Boolean(),
+                nullable=False,
+                server_default=sa.text("true"),
+            ),
+        )
+        return
+    if columns[_READ_STATE_COLUMN].get("default") is None:
+        op.alter_column(
+            _EMAIL_RECORDS_TABLE,
+            _READ_STATE_COLUMN,
+            existing_type=sa.Boolean(),
+            existing_nullable=False,
+            server_default=sa.text("true"),
+        )
+
+
+def downgrade() -> None:
+    # No-op: ``create_all`` and later model metadata may already own this
+    # column. Dropping it would remove a default this revision did not
+    # necessarily create.
+    return

backend/db/models.py

@@ -18,6 +18,7 @@
     Text,
     UniqueConstraint,
     func,
+    text,
 )
 from sqlalchemy.orm import declarative_base, Mapped, mapped_column, relationship
 from sqlalchemy.types import TypeDecorator
@@ -803,7 +804,12 @@ def owner_filters(cls, user_id: str, organization_id: str | None):
     date: Mapped[datetime.datetime] = mapped_column(DateTime(timezone=True), index=True)
     body: Mapped[str] = mapped_column(Text)
     # IMAP \Seen read state; defaults read so historical/file imports don't nag.
-    is_read: Mapped[bool] = mapped_column(Boolean, nullable=False, default=True)
+    # A DB-level server_default keeps create_all/bootstrap_db consistent with the
+    # 0011_email_read_state migration intent so raw inserts that omit is_read
+    # (e.g. postgres smoke seeds) don't hit a NOT NULL violation.
+    is_read: Mapped[bool] = mapped_column(
+        Boolean, nullable=False, default=True, server_default=text("true")
+    )
     # Defer large pgvector payloads on default entity loads.
     embedding = mapped_column(Vector(1536), deferred=True)
     attachments: Mapped[list["Attachment"]] = relationship(

backend/scripts/bootstrap_db.py

@@ -186,10 +186,6 @@ def _get_create_indexes_statements() -> list[Executable]:
             "CREATE INDEX IF NOT EXISTS ix_email_records_owner_date "
             "ON email_records (user_id, organization_id, date)"
         ),
-        text(
-            "CREATE INDEX IF NOT EXISTS ix_emails_owner_date "
-            "ON emails (user_id, organization_id, date)"
-        ),
         text(
             "CREATE INDEX IF NOT EXISTS ix_sender_relationships_owner_source "
             "ON sender_relationships "