databricks-tellr-app 0.3.13.dev3__tar.gz → 0.3.13.dev4__tar.gz

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: databricks-tellr-app
-Version: 0.3.13.dev3
+Version: 0.3.13.dev4
 Summary: Tellr application package for Databricks Apps
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
@@ -24,7 +24,7 @@ Requires-Dist: tinycss2==1.3.0
 Requires-Dist: esprima==4.0.1
 Requires-Dist: sqlalchemy==2.0.36
 Requires-Dist: psycopg2-binary==2.9.10
-Requires-Dist: databricks-sql-connector>=3.0.0
+Requires-Dist: databricks-sql-connector==3.0.1
 Requires-Dist: httpx==0.28.1
 Requires-Dist: jinja2==3.1.2
 Requires-Dist: python-dotenv==1.2.1
@@ -32,12 +32,12 @@ Requires-Dist: pyyaml==6.0.1
 Requires-Dist: pandas==2.3.3
 Requires-Dist: python-pptx==0.6.23
 Requires-Dist: playwright==1.48.0
-Requires-Dist: Pillow>=10.0.0
-Requires-Dist: google-api-python-client>=2.100.0
-Requires-Dist: google-auth-oauthlib>=1.2.0
-Requires-Dist: google-auth-httplib2>=0.2.0
-Requires-Dist: cryptography>=42.0.0
-Requires-Dist: sse-starlette>=3.0.0
+Requires-Dist: Pillow==11.3.0
+Requires-Dist: google-api-python-client==2.190.0
+Requires-Dist: google-auth-oauthlib==1.2.4
+Requires-Dist: google-auth-httplib2==0.3.0
+Requires-Dist: cryptography==46.0.3
+Requires-Dist: sse-starlette==3.0.3
 Requires-Dist: databricks-sdk==0.120.0
 Requires-Dist: mlflow[databricks]==3.14.0
 Requires-Dist: mlflow-skinny==3.14.0

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/databricks_tellr_app/run.py

@@ -62,6 +62,18 @@ def init_database(seed_databricks_defaults: bool = False) -> None:
         logger.error(f"Failed to seed defaults: {e}\n{tb}")
         raise SystemExit(1) from e
 
+    # Seed/load the Fernet master key pre-fork (SDR-4437 CRITICAL-3) so all
+    # uvicorn workers find the encryption_keys row already present.
+    logger.info("Ensuring encryption key...")
+    try:
+        from src.core.encryption import ensure_encryption_key
+        ensure_encryption_key()
+        logger.info("Encryption key ready")
+    except Exception as e:
+        tb = traceback.format_exc()
+        logger.error(f"Failed to ensure encryption key: {e}\n{tb}")
+        raise SystemExit(1) from e
+
 
 def main() -> None:
     """Start the uvicorn server."""

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/databricks_tellr_app.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: databricks-tellr-app
-Version: 0.3.13.dev3
+Version: 0.3.13.dev4
 Summary: Tellr application package for Databricks Apps
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
@@ -24,7 +24,7 @@ Requires-Dist: tinycss2==1.3.0
 Requires-Dist: esprima==4.0.1
 Requires-Dist: sqlalchemy==2.0.36
 Requires-Dist: psycopg2-binary==2.9.10
-Requires-Dist: databricks-sql-connector>=3.0.0
+Requires-Dist: databricks-sql-connector==3.0.1
 Requires-Dist: httpx==0.28.1
 Requires-Dist: jinja2==3.1.2
 Requires-Dist: python-dotenv==1.2.1
@@ -32,12 +32,12 @@ Requires-Dist: pyyaml==6.0.1
 Requires-Dist: pandas==2.3.3
 Requires-Dist: python-pptx==0.6.23
 Requires-Dist: playwright==1.48.0
-Requires-Dist: Pillow>=10.0.0
-Requires-Dist: google-api-python-client>=2.100.0
-Requires-Dist: google-auth-oauthlib>=1.2.0
-Requires-Dist: google-auth-httplib2>=0.2.0
-Requires-Dist: cryptography>=42.0.0
-Requires-Dist: sse-starlette>=3.0.0
+Requires-Dist: Pillow==11.3.0
+Requires-Dist: google-api-python-client==2.190.0
+Requires-Dist: google-auth-oauthlib==1.2.4
+Requires-Dist: google-auth-httplib2==0.3.0
+Requires-Dist: cryptography==46.0.3
+Requires-Dist: sse-starlette==3.0.3
 Requires-Dist: databricks-sdk==0.120.0
 Requires-Dist: mlflow[databricks]==3.14.0
 Requires-Dist: mlflow-skinny==3.14.0

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/databricks_tellr_app.egg-info/SOURCES.txt

@@ -76,6 +76,7 @@ setup.py
 ./src/database/__init__.py
 ./src/database/models/__init__.py
 ./src/database/models/deck_contributor.py
+./src/database/models/encryption_key.py
 ./src/database/models/feedback.py
 ./src/database/models/genie_space.py
 ./src/database/models/google_global_credentials.py
@@ -115,13 +116,6 @@ setup.py
 ./src/services/profile_service.py
 ./src/services/streaming_callback.py
 ./src/services/validator.py
-./src/services/converter_jail/__init__.py
-./src/services/converter_jail/ast_guard.py
-./src/services/converter_jail/codeprep.py
-./src/services/converter_jail/gslides_runner.py
-./src/services/converter_jail/jail.py
-./src/services/converter_jail/pptx_runner.py
-./src/services/converter_jail/protocol.py
 ./src/services/evaluation/__init__.py
 ./src/services/evaluation/llm_judge.py
 ./src/services/identity_providers/__init__.py
@@ -226,6 +220,7 @@ src/core/user_context.py
 src/database/__init__.py
 src/database/models/__init__.py
 src/database/models/deck_contributor.py
+src/database/models/encryption_key.py
 src/database/models/feedback.py
 src/database/models/genie_space.py
 src/database/models/google_global_credentials.py
@@ -265,13 +260,6 @@ src/services/pptx_prompts_defaults.py
 src/services/profile_service.py
 src/services/streaming_callback.py
 src/services/validator.py
-src/services/converter_jail/__init__.py
-src/services/converter_jail/ast_guard.py
-src/services/converter_jail/codeprep.py
-src/services/converter_jail/gslides_runner.py
-src/services/converter_jail/jail.py
-src/services/converter_jail/pptx_runner.py
-src/services/converter_jail/protocol.py
 src/services/evaluation/__init__.py
 src/services/evaluation/llm_judge.py
 src/services/identity_providers/__init__.py

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/databricks_tellr_app.egg-info/requires.txt

@@ -18,7 +18,7 @@ tinycss2==1.3.0
 esprima==4.0.1
 sqlalchemy==2.0.36
 psycopg2-binary==2.9.10
-databricks-sql-connector>=3.0.0
+databricks-sql-connector==3.0.1
 httpx==0.28.1
 jinja2==3.1.2
 python-dotenv==1.2.1
@@ -26,12 +26,12 @@ pyyaml==6.0.1
 pandas==2.3.3
 python-pptx==0.6.23
 playwright==1.48.0
-Pillow>=10.0.0
-google-api-python-client>=2.100.0
-google-auth-oauthlib>=1.2.0
-google-auth-httplib2>=0.2.0
-cryptography>=42.0.0
-sse-starlette>=3.0.0
+Pillow==11.3.0
+google-api-python-client==2.190.0
+google-auth-oauthlib==1.2.4
+google-auth-httplib2==0.3.0
+cryptography==46.0.3
+sse-starlette==3.0.3
 databricks-sdk==0.120.0
 mlflow[databricks]==3.14.0
 mlflow-skinny==3.14.0

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "databricks-tellr-app"
-version = "0.3.13.dev3"
+version = "0.3.13.dev4"
 description = "Tellr application package for Databricks Apps"
 readme = "README.md"
 requires-python = ">=3.10"
@@ -29,7 +29,7 @@ dependencies = [
     "esprima==4.0.1",
     "sqlalchemy==2.0.36",
     "psycopg2-binary==2.9.10",
-    "databricks-sql-connector>=3.0.0",
+    "databricks-sql-connector==3.0.1",
     "httpx==0.28.1",
     "jinja2==3.1.2",
     "python-dotenv==1.2.1",
@@ -37,12 +37,12 @@ dependencies = [
     "pandas==2.3.3",
     "python-pptx==0.6.23",
     "playwright==1.48.0",
-    "Pillow>=10.0.0",
-    "google-api-python-client>=2.100.0",
-    "google-auth-oauthlib>=1.2.0",
-    "google-auth-httplib2>=0.2.0",
-    "cryptography>=42.0.0",
-    "sse-starlette>=3.0.0",
+    "Pillow==11.3.0",
+    "google-api-python-client==2.190.0",
+    "google-auth-oauthlib==1.2.4",
+    "google-auth-httplib2==0.3.0",
+    "cryptography==46.0.3",
+    "sse-starlette==3.0.3",
     # --- Resolver-speedup pins: the backtracking-driver cluster ---
     # Databricks Apps installs this wheel in a platform-managed BUILD phase with a
     # ~10-minute budget. Left ranged, the databricks-langchain ->
@@ -58,6 +58,8 @@ dependencies = [
     # via a devloop deploy if these constraints change. Formerly injected at
     # deploy time as _RESOLVER_SPEEDUP_PINS; now carried in the wheel so a bare
     # `pip install databricks-tellr-app` is safe.
+    # SDR-4437 MEDIUM-7 (July 2026): the 7 formerly-ranged direct deps above
+    # this block are now exact-pinned too; same refresh discipline.
     "databricks-sdk==0.120.0",
     "mlflow[databricks]==3.14.0",
     "mlflow-skinny==3.14.0",

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/src/api/services/export_job_queue.py

@@ -307,6 +307,9 @@ async def convert_slides_with_progress(
         output_path: Path to save PPTX
         chart_images_per_slide: Chart images per slide
     """
+    from pptx import Presentation
+    from pptx.util import Inches
+
     total_slides = len(slides_html)
 
     # -- Phase 1: Prepare assets + parallel LLM code generation ---------------
@@ -333,17 +336,27 @@ async def convert_slides_with_progress(
         slide_inputs, on_codegen_progress=_on_codegen_progress,
     )
 
-    # -- Phase 2: Sequential execution INSIDE the subprocess jail -------------
-    def _relay(current: int, total: int, message: str) -> None:
-        update_export_progress(job_id, current, total, status_message=message)
+    # -- Phase 2: Sequential execution ----------------------------------------
+    prs = Presentation()
+    prs.slide_width = Inches(10)
+    prs.slide_height = Inches(7.5)
 
-    # to_thread: keep the long blocking jail wait off the event loop (this
-    # coroutine runs on the loop; _relay is fired from the jail's stdout
-    # reader thread and update_export_progress opens its own session per call).
-    await asyncio.to_thread(
-        converter._run_pptx_conversion,
-        codes, slide_inputs, output_path, progress_cb=_relay,
-    )
+    for i, (code, (html_str, _chart_files, _content_files, assets_dir)) in enumerate(
+        zip(codes, slide_inputs), 1,
+    ):
+        update_export_progress(
+            job_id, i, total_slides,
+            status_message=f"Building slide {i}/{total_slides}…",
+        )
+
+        converter._execute_slide(code, prs, html_str, assets_dir, i)
+
+        logger.debug(
+            f"Processed slide {i}/{total_slides}",
+            extra={"job_id": job_id, "slide": i, "total": total_slides},
+        )
+
+    prs.save(output_path)
 
 
 async def export_worker() -> None:

databricks_tellr_app-0.3.13.dev4/src/core/encryption.py

@@ -0,0 +1,125 @@
+"""Fernet symmetric encryption for sensitive data (OAuth credentials, tokens).
+
+Uses AES-128-CBC with HMAC-SHA256 authentication via the ``cryptography``
+library. The master key lives in the ``encryption_keys`` table of the
+application database (SDR-4437 CRITICAL-3): dynamic, persistent across
+restarts/restores, unique per deployment, zero operator setup.
+
+Resolution is SELECT-first, then race-safe INSERT-if-absent:
+
+1. ``SELECT key_value FROM encryption_keys WHERE id = 1`` — read-first is
+   load-bearing: on upgraded installs the table is created by the *deployer*
+   role and the app SP holds an explicit SELECT, INSERT grant; reading first
+   means INSERT privilege is never exercised when the row already exists.
+2. If absent, seed from a legacy ``.encryption_key`` file in the project root
+   (local dev — keeps pre-existing dev ciphertext decryptable), else generate
+   a fresh key.
+3. ``INSERT ... ON CONFLICT (id) DO NOTHING`` + read-back, so concurrent
+   workers/replicas converge on one key.
+
+There is deliberately NO env-var fallback: the deploy tool relocates the
+legacy ``GOOGLE_OAUTH_ENCRYPTION_KEY`` into this table at update time, and
+old-tool + new-wheel skew is unsupported (see the SDR-4437 remediation
+design, PR-3).
+"""
+
+import logging
+from functools import lru_cache
+from pathlib import Path
+
+from cryptography.fernet import Fernet, InvalidToken
+from sqlalchemy import text
+
+logger = logging.getLogger(__name__)
+
+_KEY_FILE = Path(__file__).resolve().parents[2] / ".encryption_key"
+
+# NOTE: these statements are deliberately schema-UNQUALIFIED, unlike the
+# _qual()-qualified raw SQL in database.py::_run_migrations. This module must
+# also run against SQLite (unit tests — no schemas) and schema-less local
+# Postgres. On Lakebase the bare name resolves to
+# <LAKEBASE_SCHEMA>.encryption_keys ONLY because _get_database_url appends
+# options=-csearch_path%3D<schema> to the URL (database.py:239,259) and the
+# do_connect listener injects just the password, preserving those options.
+# If that search_path mechanism ever changes, these statements would silently
+# target public.encryption_keys — keep this coupling in mind.
+_SELECT_KEY = text("SELECT key_value FROM encryption_keys WHERE id = 1")
+_INSERT_KEY = text(
+    "INSERT INTO encryption_keys (id, key_value, created_at) "
+    "VALUES (1, :key_value, CURRENT_TIMESTAMP) "
+    "ON CONFLICT (id) DO NOTHING"
+)
+
+
+def _validated(key: bytes) -> bytes:
+    try:
+        Fernet(key)
+    except (ValueError, TypeError) as exc:
+        raise RuntimeError(
+            "encryption_keys row id=1 is not a valid Fernet key — refusing to "
+            "start with a corrupt master key. Restore the correct key value "
+            "before restarting."
+        ) from exc
+    return key
+
+
+def _seed_value() -> bytes:
+    """Pick the value to seed an empty encryption_keys table with."""
+    if _KEY_FILE.exists():
+        key = _KEY_FILE.read_text().strip()
+        if key:
+            logger.info("Seeding encryption_keys from legacy key file %s", _KEY_FILE)
+            return key.encode()
+    logger.info("Generating new Fernet master key (fresh install)")
+    return Fernet.generate_key()
+
+
+@lru_cache(maxsize=1)
+def get_encryption_key() -> bytes:
+    """Return the Fernet master key from the encryption_keys table (id=1)."""
+    from src.core.database import get_db_session
+
+    # 1. Read-first (see module docstring for why this order matters).
+    with get_db_session() as session:
+        row = session.execute(_SELECT_KEY).first()
+    if row and row[0]:
+        return _validated(row[0].encode())
+
+    # 2 + 3. Seed if absent — race-safe across workers/replicas.
+    seed = _seed_value()
+    with get_db_session() as session:
+        session.execute(_INSERT_KEY, {"key_value": seed.decode()})
+    with get_db_session() as session:
+        row = session.execute(_SELECT_KEY).first()
+    if not row or not row[0]:
+        raise RuntimeError(
+            "Failed to seed encryption_keys: no key row after insert. "
+            "Check the app's SELECT, INSERT grants on the data schema."
+        )
+    return _validated(row[0].encode())
+
+
+def ensure_encryption_key() -> None:
+    """Pre-fork boot hook: load/seed the key before uvicorn workers fork.
+
+    Called from the deployed app's init_database step so all workers find
+    the row already present (the lazy path in get_encryption_key covers
+    local dev and replicas regardless).
+    """
+    get_encryption_key()
+
+
+def encrypt_data(plaintext: str) -> str:
+    """Encrypt *plaintext* and return a base64-encoded ciphertext string."""
+    f = Fernet(get_encryption_key())
+    return f.encrypt(plaintext.encode()).decode()
+
+
+def decrypt_data(ciphertext: str) -> str:
+    """Decrypt a base64-encoded *ciphertext* string and return the plaintext.
+
+    Raises:
+        cryptography.fernet.InvalidToken: If the key does not match or data is corrupt.
+    """
+    f = Fernet(get_encryption_key())
+    return f.decrypt(ciphertext.encode()).decode()

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/src/core/lakebase.py

@@ -280,120 +280,6 @@ def get_lakebase_connection_url(
     return url
 
 
-def setup_lakebase_schema(
-    instance_name: str,
-    schema: str,
-    client_id: str,
-    user: Optional[str] = None,
-    client: Optional[WorkspaceClient] = None,
-    endpoint_name: Optional[str] = None,
-    host: Optional[str] = None,
-    lakebase_type: str = "provisioned",
-) -> None:
-    """
-    Set up schema and grant permissions in Lakebase for an app.
-
-    Supports both provisioned and autoscaling Lakebase.
-
-    This function connects to the Lakebase instance and:
-    1. Creates the schema if it doesn't exist
-    2. Grants necessary permissions to the app's service principal (client_id)
-
-    Note: This should be called after the app is created and the database
-    resource is attached, as the client_id comes from the app's service principal.
-
-    Args:
-        instance_name: Name of the Lakebase instance
-        schema: Schema name to create
-        client_id: App's service principal client ID (becomes Postgres role)
-        user: Postgres username for connection (uses default if not provided)
-        client: Optional WorkspaceClient
-        endpoint_name: Full endpoint resource name (autoscaling only)
-        host: Pre-resolved host (autoscaling only)
-        lakebase_type: 'autoscaling' or 'provisioned'
-
-    Raises:
-        LakebaseError: If schema setup fails
-    """
-    logger.info(
-        "Setting up Lakebase schema",
-        extra={
-            "instance_name": instance_name, "schema": schema,
-            "client_id": client_id, "lakebase_type": lakebase_type,
-        },
-    )
-
-    try:
-        # Get connection info
-        conn_info = get_lakebase_connection_info(
-            instance_name=instance_name,
-            user=user,
-            client=client,
-            endpoint_name=endpoint_name,
-            host=host,
-            lakebase_type=lakebase_type,
-        )
-
-        # Use psycopg2 directly for DDL operations
-        import psycopg2
-
-        conn = psycopg2.connect(
-            host=conn_info["host"],
-            port=conn_info["port"],
-            user=conn_info["user"],
-            password=conn_info["password"],
-            dbname=conn_info["database"],
-            sslmode="require",
-        )
-        conn.autocommit = True
-
-        with conn.cursor() as cur:
-            # Create schema if not exists
-            cur.execute(f'CREATE SCHEMA IF NOT EXISTS "{schema}"')
-            logger.info(f"Schema created or verified: {schema}")
-
-            # Grant permissions to app's service principal
-            # The client_id is automatically available as a Postgres role when
-            # the database is added as an app resource
-            cur.execute(f'GRANT USAGE ON SCHEMA "{schema}" TO "{client_id}"')
-            cur.execute(f'GRANT CREATE ON SCHEMA "{schema}" TO "{client_id}"')
-
-            # Grant permissions on all tables in schema (for existing tables)
-            cur.execute(
-                f'GRANT SELECT, INSERT, UPDATE, DELETE ON ALL TABLES IN SCHEMA "{schema}" TO "{client_id}"'
-            )
-
-            # Grant permissions on all sequences in schema (for auto-increment columns)
-            cur.execute(
-                f'GRANT USAGE, SELECT ON ALL SEQUENCES IN SCHEMA "{schema}" TO "{client_id}"'
-            )
-
-            # Set default privileges for future tables
-            cur.execute(
-                f'ALTER DEFAULT PRIVILEGES IN SCHEMA "{schema}" '
-                f'GRANT SELECT, INSERT, UPDATE, DELETE ON TABLES TO "{client_id}"'
-            )
-
-            # Set default privileges for future sequences
-            cur.execute(
-                f'ALTER DEFAULT PRIVILEGES IN SCHEMA "{schema}" '
-                f'GRANT USAGE, SELECT ON SEQUENCES TO "{client_id}"'
-            )
-
-            logger.info(f"Permissions granted to {client_id} on schema {schema}")
-
-        conn.close()
-
-    except ImportError:
-        raise LakebaseError(
-            "psycopg2 is required for Lakebase schema setup. "
-            "Install with: pip install psycopg2-binary"
-        )
-    except Exception as e:
-        logger.error(f"Failed to setup Lakebase schema: {e}", exc_info=True)
-        raise LakebaseError(f"Lakebase schema setup failed: {e}") from e
-
-
 def initialize_lakebase_tables(
     instance_name: str,
     schema: str,

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/src/database/models/__init__.py

@@ -1,6 +1,7 @@
 """Database models."""
 
 from src.database.models.deck_contributor import DeckContributor
+from src.database.models.encryption_key import EncryptionKey
 from src.database.models.feedback import FeedbackConversation, SurveyResponse
 from src.database.models.genie_space import ConfigGenieSpace
 from src.database.models.google_global_credentials import GoogleGlobalCredentials
@@ -34,6 +35,7 @@ __all__ = [
     "ChatRequest",
     "ConfigGenieSpace",
     "DeckContributor",
+    "EncryptionKey",
     "ConfigProfile",
     "ConfigProfileContributor",
     "ConfigPrompts",

databricks_tellr_app-0.3.13.dev4/src/database/models/encryption_key.py

@@ -0,0 +1,30 @@
+"""Application-managed encryption key storage (SDR-4437 CRITICAL-3).
+
+Holds the Fernet master key for OAuth credential/token encryption in the
+ACL-governed Lakebase data schema instead of app.yaml. Single-row table
+(id = 1). Deliberately shares the data schema's grants: the key carries
+the same ACLs as the ciphertext it protects — an explicitly accepted risk
+in the SDR-4437 remediation design. Do NOT add key-specific grant
+tightening here.
+"""
+
+from sqlalchemy import Column, DateTime, Integer, Text
+
+from src.core.database import Base
+
+
+class EncryptionKey(Base):
+    """Fernet master key for OAuth credential/token encryption."""
+
+    __tablename__ = "encryption_keys"
+
+    id = Column(Integer, primary_key=True, autoincrement=False)
+    key_value = Column(Text, nullable=False)
+    # No Python-side default: every insert path (the app's raw-SQL seed and
+    # the deploy migration DDL) supplies CURRENT_TIMESTAMP explicitly, so a
+    # default would be dead code — and the codebase-conventional
+    # datetime.utcnow is deprecated on Python 3.12+.
+    created_at = Column(DateTime, nullable=False)
+
+    def __repr__(self) -> str:  # never print key material
+        return f"<EncryptionKey(id={self.id}, created_at={self.created_at})>"

{databricks_tellr_app-0.3.13.dev3 → databricks_tellr_app-0.3.13.dev4}/src/services/google_slides_prompts_defaults.py

@@ -69,9 +69,8 @@ TWO-COLUMN LAYOUT (.two-column):
 - Left: left=0.4", width=4.3"; Right: left=5.0", width=4.5"; Gap: 0.3"
 
 TABLES (<table>):
-- createTable first (rows, cols, size, position), with an explicit objectId
-- AFTER it in the returned list: insertText per cell + updateTextStyle + updateTableCellProperties
-  (the host executes requests in order, so the table exists before its cells are filled)
+- createTable in FIRST batchUpdate (rows, cols, size, position)
+- Then insertText per cell + updateTextStyle + updateTableCellProperties in SECOND batchUpdate
 - updateTableCellProperties requires: 'tableRange': {'location': {'rowIndex': r, 'columnIndex': c}, 'rowSpan': 1, 'columnSpan': 1}
 - Headers: background from CSS (typically dark), text white, font 8pt bold
 - Cells: font 7-8pt, preserve colors, left-align text, right-align numbers
@@ -108,22 +107,17 @@ REGULAR SLIDES:
   Subtitle must fit in 0.28" (10pt ≈ 0.14" per line → 2 lines max).
 
 CHARTS (CRITICAL):
-- If chart images are listed in the user message, you MUST reference them by
-  placeholder — do NOT upload anything, do NOT call any Google API.
-- For each chart image `filename` in assets_dir, add a createImage request with
-  url set to the placeholder 'tellr-asset://' + filename. The host uploads the
-  file to Drive and substitutes the real URL before execution.
-  Example: {'createImage': {'objectId': 'img_' + str(uuid.uuid4())[:8],
-    'url': 'tellr-asset://' + filename,
-    'elementProperties': {'pageObjectId': page_id, 'size': {...}, 'transform': {...}}}}
+- If chart images are listed in the user message, you MUST use them as pre-captured screenshots.
 - Do NOT recreate charts with matplotlib, plotly, or any library. The images are already rendered.
+- Upload to Drive via MediaFileUpload, set anyone/reader permission, then createImage.
 - Chart only (no metric cards): left=0.4", top=1.1", width=9.0", height=4.0" — fills the body zone.
 - Chart + metrics: use CHART + METRICS SIDE-BY-SIDE layout above.
 - If NO chart images are listed, skip chart rendering entirely — do not attempt to recreate.
 
 CONTENT IMAGES (logos, icons — CRITICAL):
-- Reference each content image the same way: url = 'tellr-asset://' + filename.
-- NEVER upload; NEVER create Drive files or set permissions yourself. The host does that.
+- If content images are listed in the user message, you MUST upload and add them to the slide.
+- These are pre-extracted image files (PNG) saved in assets_dir — ready to upload.
+- Upload to Drive the same way as charts: MediaFileUpload → files().create → permissions().create → createImage.
 - Position to match the HTML layout (e.g. a logo in the top-right header area → right-aligned near the title).
 - Typical logo sizes: width=0.6"-1.0", height=0.3"-0.5". Do NOT make logos larger than needed.
 - Place logos/icons in the header zone (top=0.15"-0.4") so they don't overlap body content.
@@ -171,27 +165,23 @@ API PATTERNS (CRITICAL — wrong nesting causes 400 errors):
 
 ALLOWED IMPORTS:
 import os, json, uuid, re
-(Do NOT import googleapiclient — you never touch the network.)
+from googleapiclient.http import MediaFileUpload  # ONLY when uploading images
 
 EXECUTION:
 - ALL code MUST be inside the function body. Nothing at module level except imports and helpers.
-- Build ONE requests list and RETURN it. Never call batchUpdate or any API
-  yourself — the host executes the returned requests in order.
-- Do NOT wrap request-building in try/except.
+- Build requests list, then ONE batchUpdate (except tables need two).
+- Do NOT wrap batchUpdate in try/except.
 - Every createShape needs a unique objectId: 'txt_' + str(uuid.uuid4())[:8]
 
 Return ONLY Python code, no markdown fences."""
 
 # ── Multi-slide prompts ──────────────────────────────────────────────
 
-DEFAULT_GSLIDES_SYSTEM_PROMPT = """Generate Python code with a build_slide_requests(html_str, assets_dir, page_id) function that RETURNS a list of Google Slides batchUpdate request dicts.
+DEFAULT_GSLIDES_SYSTEM_PROMPT = """Generate Python code with add_slide_to_presentation(slides_service, drive_service, presentation_id, page_id, html_str, assets_dir) function.
 
 CONTEXT:
-- You do NOT have and do NOT need any Google service objects or network access.
-- Do NOT call batchUpdate or any other Google API method — never execute requests, upload files, or create permissions yourself.
-- The target slide page already exists; reference it via the page_id argument.
-- Return (do not execute) a Python list of request dicts. The trusted host
-  executes them, uploads any referenced images, and handles chunking/retries.
+- All arguments are pre-created. Do NOT import googleapiclient or build services.
+- The slide page already exists (page_id). Do NOT call createSlide — only add content.
 
 HELPERS (include in your code):
 def emu(inches): return int(inches * 914400)
@@ -202,22 +192,22 @@ def hex_to_rgb(h):
 SLIDE: 16:9 widescreen — 10" × 5.625" (9144000 × 5143500 EMU).
 Bounds: left ≥ 0.4", top ≥ 0.2", left + width ≤ 9.6", top + height ≤ 5.4"
 
-ELEMENT CREATION (append each request dict to a list, then `return requests`):
-- TEXT_BOX: createShape + insertText + updateTextStyle (textRange {'type':'ALL'}) + updateParagraphStyle
+ELEMENT CREATION:
+- TEXT_BOX: createShape + insertText + updateTextStyle (textRange: {'type':'ALL'}) + updateParagraphStyle
 - RECTANGLE/ROUND_RECTANGLE: createShape + updateShapeProperties
 - Size: {'width': {'magnitude': emu(w), 'unit': 'EMU'}, 'height': {'magnitude': emu(h), 'unit': 'EMU'}}
 - Position: {'scaleX':1,'scaleY':1,'translateX':emu(left),'translateY':emu(top),'unit':'EMU'}
 
-The code MUST define build_slide_requests(html_str, assets_dir, page_id) and RETURN the requests list.
+The code MUST define the function add_slide_to_presentation.
 """ + _GSLIDES_SHARED_RULES
 
-DEFAULT_GSLIDES_USER_PROMPT = """Convert the HTML below to a list of Google Slides batchUpdate requests.
+DEFAULT_GSLIDES_USER_PROMPT = """Convert the HTML below to a Google Slide.
 
 {html_content}
 
 {screenshot_note}
 
-Return Python code defining build_slide_requests(html_str, assets_dir, page_id) that returns the requests list."""
+Return Python code with add_slide_to_presentation(slides_service, drive_service, presentation_id, page_id, html_str, assets_dir)."""
 
 # ── Single-slide prompts ─────────────────────────────────────────────