docs: finalize v0.3.0 changelog for release

Consolidated all unreleased changes into v0.3.0 release dated 2025-11-03.

Key additions:
- Development Mode with mock AI provider
- Config Override System for Docker
- Async Price Download (non-blocking)
- Resume Mode (idempotent execution)
- Reasoning Logs API (GET /reasoning)
- Project rebrand to AI-Trader-Server

Includes comprehensive bug fixes for context injection, simulation
re-runs, database reliability, and configuration handling.

CHANGELOG.md

@@ -7,13 +7,43 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
-### Fixed
-- **Dev Mode Warning in Docker** - DEV mode startup warning now displays correctly in Docker logs
-  - Added FastAPI `@app.on_event("startup")` handler to trigger warning on API server startup
-  - Previously only appeared when running `python api/main.py` directly (not via uvicorn)
-  - Docker compose now includes `DEPLOYMENT_MODE` and `PRESERVE_DEV_DATA` environment variables
+## [0.3.0] - 2025-11-03
 
-## [0.3.0] - 2025-10-31
+### Added - Development & Testing Features
+- **Development Mode** - Mock AI provider for cost-free testing
+  - `DEPLOYMENT_MODE=DEV` enables mock AI responses with deterministic stock rotation
+  - Isolated dev database (`trading_dev.db`) separate from production data
+  - `PRESERVE_DEV_DATA=true` option to prevent dev database reset on startup
+  - No AI API costs during development and testing
+  - All API responses include `deployment_mode` field
+  - Startup warning displayed when running in DEV mode
+- **Config Override System** - Docker configuration merging
+  - Place custom configs in `user-configs/` directory
+  - Startup merges user config with default config
+  - Comprehensive validation with clear error messages
+  - Volume mount: `./user-configs:/app/user-configs`
+
+### Added - Enhanced API Features
+- **Async Price Download** - Non-blocking data preparation
+  - `POST /simulate/trigger` no longer blocks on price downloads
+  - New job status: `downloading_data` during data preparation
+  - Warnings field in status response for download issues
+  - Better user experience for large date ranges
+- **Resume Mode** - Idempotent simulation execution
+  - Jobs automatically skip already-completed model-days
+  - Safe to re-run jobs without duplicating work
+  - `status="skipped"` for already-completed executions
+  - Error-free job completion when partial results exist
+- **Reasoning Logs API** - Access AI decision-making history
+  - `GET /reasoning` endpoint for querying reasoning logs
+  - Filter by job_id, model_name, date, include_full_conversation
+  - Includes conversation history and tool usage
+  - Database-only storage (no JSONL files)
+  - AI-powered summary generation for reasoning sessions
+- **Job Skip Status** - Enhanced job status tracking
+  - New status: `skipped` for already-completed model-days
+  - Better differentiation between pending, running, and skipped
+  - Accurate job completion detection
 
 ### Added - Price Data Management & On-Demand Downloads
 - **SQLite Price Data Storage** - Replaced JSONL files with relational database
@@ -83,13 +113,22 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - Windmill integration patterns and examples
 
 ### Changed
+- **Project Rebrand** - AI-Trader renamed to AI-Trader-Server
+  - Updated all documentation for new project name
+  - Updated Docker images to ghcr.io/xe138/ai-trader-server
+  - Updated GitHub Actions workflows
+  - Updated README, CHANGELOG, and all user guides
 - **Architecture** - Transformed from batch-only to API-first service with database persistence
 - **Data Storage** - Migrated from JSONL files to SQLite relational database
   - Price data now stored in `price_data` table instead of `merged.jsonl`
   - Tools/price_tools.py updated to query database
-  - Position data remains in database (already migrated in earlier versions)
+  - Position data fully migrated to database-only storage (removed JSONL dependencies)
+  - Trade tools now read/write from database tables with lazy context injection
 - **Deployment** - Simplified to single API-only Docker service (REST API is new in v0.3.0)
+- **Logging** - Removed duplicate MCP service log files for cleaner output
 - **Configuration** - Simplified environment variable configuration
+  - **Added:** `DEPLOYMENT_MODE` (PROD/DEV) for environment control
+  - **Added:** `PRESERVE_DEV_DATA` (default: false) to keep dev data between runs
   - **Added:** `AUTO_DOWNLOAD_PRICE_DATA` (default: true) - Enable on-demand downloads
   - **Added:** `MAX_SIMULATION_DAYS` (default: 30) - Maximum date range size
   - **Added:** `API_PORT` for host port mapping (default: 8080, customizable for port conflicts)
@@ -137,6 +176,35 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Monitoring** - Health checks and status tracking
 - **Persistence** - SQLite database survives container restarts
 
+### Fixed
+- **Context Injection** - Runtime parameters correctly injected into MCP tools
+  - ContextInjector always overrides AI-provided parameters (defense-in-depth)
+  - Hidden context parameters from AI tool schema to prevent hallucination
+  - Resolved database locking issues with concurrent tool calls
+  - Proper async handling of tool reloading after context injection
+- **Simulation Re-runs** - Prevent duplicate execution of completed model-days
+  - Fixed job hanging when re-running partially completed simulations
+  - `_execute_date()` now skips already-completed model-days
+  - Job completion status correctly reflects skipped items
+- **Agent Initialization** - Correct parameter passing in API mode
+  - Fixed BaseAgent initialization parameters in ModelDayExecutor
+  - Resolved async execution and position storage issues
+- **Database Reliability** - Various improvements for concurrent access
+  - Fixed column existence checks before creating indexes
+  - Proper database path resolution in dev mode (prevents recursive _dev suffix)
+  - Module-level database initialization for uvicorn reliability
+  - Fixed database locking during concurrent writes
+  - Improved error handling in buy/sell functions
+- **Configuration** - Improved config handling
+  - Use enabled field from config to determine which models run
+  - Use config models when empty models list provided
+  - Correct handling of merged runtime configs in containers
+  - Proper get_db_path() usage to pass base database path
+- **Docker** - Various deployment improvements
+  - Removed non-existent data scripts from Dockerfile
+  - Proper respect for dev mode in entrypoint database initialization
+  - Correct closure usage to capture db_path in lifespan context manager
+
 ### Breaking Changes
 - **Batch Mode Removed** - All simulations now run through REST API
   - v0.2.0 used sequential batch execution via Docker entrypoint
@@ -147,7 +215,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   - `merged.jsonl` no longer used (replaced by `price_data` table)
   - Automatic on-demand downloads eliminate need for manual data fetching
 - **Configuration Variables Changed**
-  - Added: `AUTO_DOWNLOAD_PRICE_DATA`, `MAX_SIMULATION_DAYS`, `API_PORT`
+  - Added: `DEPLOYMENT_MODE`, `PRESERVE_DEV_DATA`, `AUTO_DOWNLOAD_PRICE_DATA`, `MAX_SIMULATION_DAYS`, `API_PORT`
   - Removed: `RUNTIME_ENV_PATH`, MCP service ports, `WEB_HTTP_PORT`
   - MCP services now use fixed internal ports (not exposed to host)
 

agent/base_agent/base_agent.py

@@ -173,21 +173,13 @@ class BaseAgent:
                 print("⚠️  OpenAI base URL not set, using default")
 
         try:
-            # Get job_id from runtime config if available (API mode)
-            from tools.general_tools import get_config_value
-            job_id = get_config_value("JOB_ID")  # Returns None if not in API mode
+            # Context injector will be set later via set_context() method
+            self.context_injector = None
 
-            # Create context injector for injecting signature and today_date into tool calls
-            self.context_injector = ContextInjector(
-                signature=self.signature,
-                today_date=self.init_date,  # Will be updated per trading session
-                job_id=job_id  # Will be None in standalone mode, populated in API mode
-            )
-
-            # Create MCP client with interceptor
+            # Create MCP client without interceptors initially
             self.client = MultiServerMCPClient(
                 self.mcp_config,
-                tool_interceptors=[self.context_injector]
+                tool_interceptors=[]
             )
 
             # Get tools
@@ -229,6 +221,40 @@ class BaseAgent:
 
         print(f"✅ Agent {self.signature} initialization completed")
 
+    async def set_context(self, context_injector: "ContextInjector") -> None:
+        """
+        Inject ContextInjector after initialization.
+
+        This allows the ContextInjector to be created with the correct
+        trading day date and session_id after the agent is initialized.
+
+        Args:
+            context_injector: Configured ContextInjector instance with
+                            correct signature, today_date, job_id, session_id
+        """
+        print(f"[DEBUG] set_context() ENTRY: Received context_injector with signature={context_injector.signature}, date={context_injector.today_date}, job_id={context_injector.job_id}, session_id={context_injector.session_id}")
+
+        self.context_injector = context_injector
+        print(f"[DEBUG] set_context(): Set self.context_injector, id={id(self.context_injector)}")
+
+        # Recreate MCP client with the interceptor
+        # Note: We need to recreate because MultiServerMCPClient doesn't have add_interceptor()
+        print(f"[DEBUG] set_context(): Creating new MCP client with interceptor, id={id(context_injector)}")
+        self.client = MultiServerMCPClient(
+            self.mcp_config,
+            tool_interceptors=[context_injector]
+        )
+        print(f"[DEBUG] set_context(): MCP client created")
+
+        # CRITICAL: Reload tools from new client so they use the interceptor
+        print(f"[DEBUG] set_context(): Reloading tools...")
+        self.tools = await self.client.get_tools()
+        print(f"[DEBUG] set_context(): Tools reloaded, count={len(self.tools)}")
+
+        print(f"✅ Context injected: signature={context_injector.signature}, "
+              f"date={context_injector.today_date}, job_id={context_injector.job_id}, "
+              f"session_id={context_injector.session_id}")
+
     def _capture_message(self, role: str, content: str, tool_name: str = None, tool_input: str = None) -> None:
         """
         Capture a message in conversation history.
@@ -429,18 +455,32 @@ Summary:"""
         await self._handle_trading_result(today_date)
     
     async def _handle_trading_result(self, today_date: str) -> None:
-        """Handle trading results"""
+        """Handle trading results with database writes."""
+        from tools.price_tools import add_no_trade_record_to_db
+
         if_trade = get_config_value("IF_TRADE")
+
         if if_trade:
             write_config_value("IF_TRADE", False)
             print("✅ Trading completed")
         else:
             print("📊 No trading, maintaining positions")
-            try:
-                add_no_trade_record(today_date, self.signature)
-            except NameError as e:
-                print(f"❌ NameError: {e}")
-                raise
+
+            # Get context from runtime config
+            job_id = get_config_value("JOB_ID")
+            session_id = self.context_injector.session_id if self.context_injector else None
+
+            if not job_id or not session_id:
+                raise ValueError("Missing JOB_ID or session_id for no-trade record")
+
+            # Write no-trade record to database
+            add_no_trade_record_to_db(
+                today_date,
+                self.signature,
+                job_id,
+                session_id
+            )
+
             write_config_value("IF_TRADE", False)
     
     def register_agent(self) -> None:

agent/context_injector.py

@@ -49,14 +49,16 @@ class ContextInjector:
         """
         # Inject context parameters for trade tools
         if request.name in ["buy", "sell"]:
-            # Add signature and today_date to args if not present
-            if "signature" not in request.args:
-                request.args["signature"] = self.signature
-            if "today_date" not in request.args:
-                request.args["today_date"] = self.today_date
-            if "job_id" not in request.args and self.job_id:
+            # Debug: Log self attributes BEFORE injection
+            print(f"[ContextInjector.__call__] ENTRY: id={id(self)}, self.signature={self.signature}, self.today_date={self.today_date}, self.job_id={self.job_id}, self.session_id={self.session_id}")
+            print(f"[ContextInjector.__call__] Args BEFORE injection: {request.args}")
+
+            # ALWAYS inject/override context parameters (don't trust AI-provided values)
+            request.args["signature"] = self.signature
+            request.args["today_date"] = self.today_date
+            if self.job_id:
                 request.args["job_id"] = self.job_id
-            if "session_id" not in request.args and self.session_id:
+            if self.session_id:
                 request.args["session_id"] = self.session_id
 
             # Debug logging

agent_tools/tool_trade.py

@@ -7,7 +7,6 @@ project_root = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 sys.path.insert(0, project_root)
 from tools.price_tools import get_open_prices
 import json
-from tools.deployment_config import get_db_path
 from api.database import get_db_connection
 from datetime import datetime
 mcp = FastMCP("TradeTools")
@@ -30,7 +29,7 @@ def get_current_position_from_db(job_id: str, model: str, date: str) -> Tuple[Di
     Raises:
         Exception: If database query fails
     """
-    db_path = get_db_path()
+    db_path = "data/jobs.db"
     conn = get_db_connection(db_path)
     cursor = conn.cursor()
 
@@ -83,24 +82,13 @@ def get_current_position_from_db(job_id: str, model: str, date: str) -> Tuple[Di
         conn.close()
 
 
-@mcp.tool()
-def buy(symbol: str, amount: int, signature: str = None, today_date: str = None,
-        job_id: str = None, session_id: int = None) -> Dict[str, Any]:
+def _buy_impl(symbol: str, amount: int, signature: str = None, today_date: str = None,
+              job_id: str = None, session_id: int = None) -> Dict[str, Any]:
     """
-    Buy stock function - writes to SQLite database.
+    Internal buy implementation - accepts injected context parameters.
 
-    Args:
-        symbol: Stock symbol (e.g., "AAPL", "MSFT")
-        amount: Number of shares to buy (positive integer)
-        signature: Model signature (injected by ContextInjector)
-        today_date: Trading date YYYY-MM-DD (injected by ContextInjector)
-        job_id: Job UUID (injected by ContextInjector)
-        session_id: Trading session ID (injected by ContextInjector)
-
-    Returns:
-        Dict[str, Any]:
-          - Success: {"CASH": amount, symbol: quantity, ...}
-          - Failure: {"error": message, ...}
+    This function is not exposed to the AI model. It receives runtime context
+    (signature, today_date, job_id, session_id) from the ContextInjector.
     """
     # Validate required parameters
     if not job_id:
@@ -110,7 +98,7 @@ def buy(symbol: str, amount: int, signature: str = None, today_date: str = None,
     if not today_date:
         return {"error": "Missing required parameter: today_date"}
 
-    db_path = get_db_path()
+    db_path = "data/jobs.db"
     conn = get_db_connection(db_path)
     cursor = conn.cursor()
 
@@ -207,8 +195,29 @@ def buy(symbol: str, amount: int, signature: str = None, today_date: str = None,
 
 
 @mcp.tool()
-def sell(symbol: str, amount: int, signature: str = None, today_date: str = None,
-         job_id: str = None, session_id: int = None) -> Dict[str, Any]:
+def buy(symbol: str, amount: int, signature: str = None, today_date: str = None,
+        job_id: str = None, session_id: int = None) -> Dict[str, Any]:
+    """
+    Buy stock shares.
+
+    Args:
+        symbol: Stock symbol (e.g., "AAPL", "MSFT", "GOOGL")
+        amount: Number of shares to buy (positive integer)
+
+    Returns:
+        Dict[str, Any]:
+          - Success: {"CASH": remaining_cash, "SYMBOL": shares, ...}
+          - Failure: {"error": error_message, ...}
+
+    Note: signature, today_date, job_id, session_id are automatically injected by the system.
+    Do not provide these parameters - they will be added automatically.
+    """
+    # Delegate to internal implementation
+    return _buy_impl(symbol, amount, signature, today_date, job_id, session_id)
+
+
+def _sell_impl(symbol: str, amount: int, signature: str = None, today_date: str = None,
+               job_id: str = None, session_id: int = None) -> Dict[str, Any]:
     """
     Sell stock function - writes to SQLite database.
 
@@ -233,7 +242,7 @@ def sell(symbol: str, amount: int, signature: str = None, today_date: str = None
     if not today_date:
         return {"error": "Missing required parameter: today_date"}
 
-    db_path = get_db_path()
+    db_path = "data/jobs.db"
     conn = get_db_connection(db_path)
     cursor = conn.cursor()
 
@@ -328,6 +337,28 @@ def sell(symbol: str, amount: int, signature: str = None, today_date: str = None
         conn.close()
 
 
+@mcp.tool()
+def sell(symbol: str, amount: int, signature: str = None, today_date: str = None,
+         job_id: str = None, session_id: int = None) -> Dict[str, Any]:
+    """
+    Sell stock shares.
+
+    Args:
+        symbol: Stock symbol (e.g., "AAPL", "MSFT", "GOOGL")
+        amount: Number of shares to sell (positive integer)
+
+    Returns:
+        Dict[str, Any]:
+          - Success: {"CASH": remaining_cash, "SYMBOL": shares, ...}
+          - Failure: {"error": error_message, ...}
+
+    Note: signature, today_date, job_id, session_id are automatically injected by the system.
+    Do not provide these parameters - they will be added automatically.
+    """
+    # Delegate to internal implementation
+    return _sell_impl(symbol, amount, signature, today_date, job_id, session_id)
+
+
 if __name__ == "__main__":
     port = int(os.getenv("TRADE_HTTP_PORT", "8002"))
     mcp.run(transport="streamable-http", port=port)

api/model_day_executor.py

@@ -129,12 +129,21 @@ class ModelDayExecutor:
             # Set environment variable for agent to use isolated config
             os.environ["RUNTIME_ENV_PATH"] = self.runtime_config_path
 
-            # Initialize agent
+            # Initialize agent (without context)
             agent = await self._initialize_agent()
 
-            # Update context injector with session_id
-            if hasattr(agent, 'context_injector') and agent.context_injector:
-                agent.context_injector.session_id = session_id
+            # Create and inject context with correct values
+            from agent.context_injector import ContextInjector
+            context_injector = ContextInjector(
+                signature=self.model_sig,
+                today_date=self.date,  # Current trading day
+                job_id=self.job_id,
+                session_id=session_id
+            )
+            logger.info(f"[DEBUG] ModelDayExecutor: Created ContextInjector with signature={self.model_sig}, date={self.date}, job_id={self.job_id}, session_id={session_id}")
+            logger.info(f"[DEBUG] ModelDayExecutor: Calling await agent.set_context()")
+            await agent.set_context(context_injector)
+            logger.info(f"[DEBUG] ModelDayExecutor: set_context() completed")
 
             # Run trading session
             logger.info(f"Running trading session for {self.model_sig} on {self.date}")
@@ -149,10 +158,13 @@ class ModelDayExecutor:
             # Update session summary
             await self._update_session_summary(cursor, session_id, conversation, agent)
 
-            # Store positions (pass session_id)
-            self._write_results_to_db(agent, session_id)
-
+            # Commit and close connection before _write_results_to_db opens a new one
             conn.commit()
+            conn.close()
+            conn = None  # Mark as closed
+
+            # Store positions (pass session_id) - this opens its own connection
+            self._write_results_to_db(agent, session_id)
 
             # Update status to completed
             self.job_manager.update_job_detail_status(

api/simulation_worker.py

@@ -90,7 +90,7 @@ class SimulationWorker:
             logger.info(f"Starting job {self.job_id}: {len(date_range)} dates, {len(models)} models")
 
             # NEW: Prepare price data (download if needed)
-            available_dates, warnings = self._prepare_data(date_range, models, config_path)
+            available_dates, warnings, completion_skips = self._prepare_data(date_range, models, config_path)
 
             if not available_dates:
                 error_msg = "No trading dates available after price data preparation"
@@ -100,7 +100,7 @@ class SimulationWorker:
             # Execute available dates only
             for date in available_dates:
                 logger.info(f"Processing date {date} with {len(models)} models")
-                self._execute_date(date, models, config_path)
+                self._execute_date(date, models, config_path, completion_skips)
 
             # Job completed - determine final status
             progress = self.job_manager.get_job_progress(self.job_id)
@@ -145,7 +145,8 @@ class SimulationWorker:
                 "error": error_msg
             }
 
-    def _execute_date(self, date: str, models: List[str], config_path: str) -> None:
+    def _execute_date(self, date: str, models: List[str], config_path: str,
+                      completion_skips: Dict[str, Set[str]] = None) -> None:
         """
         Execute all models for a single date in parallel.
 
@@ -153,14 +154,24 @@ class SimulationWorker:
             date: Trading date (YYYY-MM-DD)
             models: List of model signatures to execute
             config_path: Path to configuration file
+            completion_skips: {model: {dates}} of already-completed model-days to skip
 
         Uses ThreadPoolExecutor to run all models concurrently for this date.
         Waits for all models to complete before returning.
+        Skips models that have already completed this date.
         """
+        if completion_skips is None:
+            completion_skips = {}
+
         with ThreadPoolExecutor(max_workers=self.max_workers) as executor:
             # Submit all model executions for this date
             futures = []
             for model in models:
+                # Skip if this model-day was already completed
+                if date in completion_skips.get(model, set()):
+                    logger.debug(f"Skipping {model} on {date} (already completed)")
+                    continue
+
                 future = executor.submit(
                     self._execute_model_day,
                     date,
@@ -397,7 +408,10 @@ class SimulationWorker:
             config_path: Path to configuration file
 
         Returns:
-            Tuple of (available_dates, warnings)
+            Tuple of (available_dates, warnings, completion_skips)
+            - available_dates: Dates to process
+            - warnings: Warning messages
+            - completion_skips: {model: {dates}} of already-completed model-days
         """
         from api.price_data_manager import PriceDataManager
 
@@ -456,7 +470,7 @@ class SimulationWorker:
         self.job_manager.update_job_status(self.job_id, "running")
         logger.info(f"Job {self.job_id}: Starting execution - {len(dates_to_process)} dates, {len(models)} models")
 
-        return dates_to_process, warnings
+        return dates_to_process, warnings, completion_skips
 
     def get_job_info(self) -> Dict[str, Any]:
         """

docs/plans/2025-02-11-database-position-tracking-design.md

@@ -0,0 +1,476 @@
+# Database-Only Position Tracking Design
+
+**Date:** 2025-02-11
+**Status:** Approved
+**Version:** 1.0
+
+## Problem Statement
+
+Two critical issues prevent simulations from running:
+
+1. **ContextInjector receives None values**: The ContextInjector shows `{'signature': None, 'today_date': None, 'job_id': None, 'session_id': None}` when injecting parameters into trade tool calls, causing trade validation to fail.
+
+2. **File-based position tracking still in use**: System prompt builder and no-trade handler attempt to read/write position.jsonl files that no longer exist after SQLite migration.
+
+## Root Cause Analysis
+
+### Issue 1: ContextInjector Initialization Timing
+
+**Problem Chain:**
+- `BaseAgent.__init__()` creates `ContextInjector` with `self.init_date`
+- `init_date` is the START of simulation date range (e.g., "2025-10-13"), not current trading day ("2025-10-01")
+- Runtime config contains correct values (`TODAY_DATE="2025-10-01"`, `SIGNATURE="gpt-5"`, `JOB_ID="dc488e87..."`), but BaseAgent doesn't use them during initialization
+- ContextInjector is created before the trading session, so it doesn't know the correct date
+
+**Evidence:**
+```
+ai-trader-app  | [ContextInjector] Tool: buy, Args after injection: {'symbol': 'MSFT', 'amount': 1, 'signature': None, 'today_date': None, 'job_id': None, 'session_id': None}
+```
+
+### Issue 2: Mixed Storage Architecture
+
+**Problem Chain:**
+- Trade tools (tool_trade.py) query/write to SQLite database
+- System prompt builder calls `get_today_init_position()` which reads position.jsonl files
+- No-trade handler calls `add_no_trade_record()` which writes to position.jsonl files
+- Files don't exist because we migrated to database-only storage
+
+**Evidence:**
+```
+FileNotFoundError: [Errno 2] No such file or directory: '/app/data/agent_data/gpt-5/position/position.jsonl'
+```
+
+## Design Solution
+
+### Architecture Principles
+
+1. **Database-only position storage**: All position queries and writes go through SQLite
+2. **Lazy context injection**: Create ContextInjector after runtime config is written and session is created
+3. **Real-time database queries**: System prompt builder queries database directly, no file caching
+4. **Clean initialization order**: Config → Database → Agent → Context → Session
+
+### Component Changes
+
+#### 1. ContextInjector Lifecycle Refactor
+
+**BaseAgent Changes:**
+
+Remove ContextInjector creation from `__init__()`:
+```python
+# OLD (in __init__)
+self.context_injector = ContextInjector(
+    signature=self.signature,
+    today_date=self.init_date,  # WRONG: uses start date
+    job_id=job_id
+)
+self.client = MultiServerMCPClient(
+    self.mcp_config,
+    tool_interceptors=[self.context_injector]
+)
+
+# NEW (in __init__)
+self.context_injector = None
+self.client = MultiServerMCPClient(
+    self.mcp_config,
+    tool_interceptors=[]  # Empty initially
+)
+```
+
+Add new method `set_context()`:
+```python
+def set_context(self, context_injector: ContextInjector) -> None:
+    """Inject ContextInjector after initialization.
+
+    Args:
+        context_injector: Configured ContextInjector instance
+    """
+    self.context_injector = context_injector
+    self.client.add_interceptor(context_injector)
+```
+
+**ModelDayExecutor Changes:**
+
+Create and inject ContextInjector after agent initialization:
+```python
+async def execute_async(self) -> Dict[str, Any]:
+    # ... create session, initialize position ...
+
+    # Set RUNTIME_ENV_PATH
+    os.environ["RUNTIME_ENV_PATH"] = self.runtime_config_path
+
+    # Initialize agent (without context)
+    agent = await self._initialize_agent()
+
+    # Create context injector with correct values
+    context_injector = ContextInjector(
+        signature=self.model_sig,
+        today_date=self.date,  # CORRECT: current trading day
+        job_id=self.job_id,
+        session_id=session_id
+    )
+
+    # Inject context into agent
+    agent.set_context(context_injector)
+
+    # Run trading session
+    session_result = await agent.run_trading_session(self.date)
+```
+
+#### 2. Database Position Query Functions
+
+**New Functions (tools/price_tools.py):**
+
+```python
+def get_today_init_position_from_db(
+    today_date: str,
+    modelname: str,
+    job_id: str
+) -> Dict[str, float]:
+    """
+    Query yesterday's position from database.
+
+    Args:
+        today_date: Current trading date (YYYY-MM-DD)
+        modelname: Model signature
+        job_id: Job UUID
+
+    Returns:
+        Position dict: {"AAPL": 50, "MSFT": 30, "CASH": 5000.0}
+        If no position exists: {"CASH": 10000.0} (initial cash)
+    """
+    from tools.deployment_config import get_db_path
+    from api.database import get_db_connection
+
+    db_path = get_db_path()
+    conn = get_db_connection(db_path)
+    cursor = conn.cursor()
+
+    try:
+        # Get most recent position before today
+        cursor.execute("""
+            SELECT p.id, p.cash
+            FROM positions p
+            WHERE p.job_id = ? AND p.model = ? AND p.date < ?
+            ORDER BY p.date DESC, p.action_id DESC
+            LIMIT 1
+        """, (job_id, modelname, today_date))
+
+        row = cursor.fetchone()
+
+        if not row:
+            # First day - return initial cash
+            return {"CASH": 10000.0}  # TODO: Read from config
+
+        position_id, cash = row
+        position_dict = {"CASH": cash}
+
+        # Get holdings for this position
+        cursor.execute("""
+            SELECT symbol, quantity
+            FROM holdings
+            WHERE position_id = ?
+        """, (position_id,))
+
+        for symbol, quantity in cursor.fetchall():
+            position_dict[symbol] = quantity
+
+        return position_dict
+
+    finally:
+        conn.close()
+
+
+def add_no_trade_record_to_db(
+    today_date: str,
+    modelname: str,
+    job_id: str,
+    session_id: int
+) -> None:
+    """
+    Create no-trade position record in database.
+
+    Args:
+        today_date: Current trading date (YYYY-MM-DD)
+        modelname: Model signature
+        job_id: Job UUID
+        session_id: Trading session ID
+    """
+    from tools.deployment_config import get_db_path
+    from api.database import get_db_connection
+    from agent_tools.tool_trade import get_current_position_from_db
+    from datetime import datetime
+
+    db_path = get_db_path()
+    conn = get_db_connection(db_path)
+    cursor = conn.cursor()
+
+    try:
+        # Get current position
+        current_position, next_action_id = get_current_position_from_db(
+            job_id, modelname, today_date
+        )
+
+        # Calculate portfolio value
+        # (Reuse logic from tool_trade.py)
+        cash = current_position.get("CASH", 0.0)
+        portfolio_value = cash
+
+        # Add stock values
+        for symbol, qty in current_position.items():
+            if symbol != "CASH":
+                try:
+                    from tools.price_tools import get_open_prices
+                    price = get_open_prices(today_date, [symbol])[f'{symbol}_price']
+                    portfolio_value += qty * price
+                except KeyError:
+                    pass
+
+        # Get previous value for P&L
+        cursor.execute("""
+            SELECT portfolio_value
+            FROM positions
+            WHERE job_id = ? AND model = ? AND date < ?
+            ORDER BY date DESC, action_id DESC
+            LIMIT 1
+        """, (job_id, modelname, today_date))
+
+        row = cursor.fetchone()
+        previous_value = row[0] if row else 10000.0
+
+        daily_profit = portfolio_value - previous_value
+        daily_return_pct = (daily_profit / previous_value * 100) if previous_value > 0 else 0
+
+        # Insert position record
+        created_at = datetime.utcnow().isoformat() + "Z"
+
+        cursor.execute("""
+            INSERT INTO positions (
+                job_id, date, model, action_id, action_type,
+                cash, portfolio_value, daily_profit, daily_return_pct,
+                session_id, created_at
+            )
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        """, (
+            job_id, today_date, modelname, next_action_id, "no_trade",
+            cash, portfolio_value, daily_profit, daily_return_pct,
+            session_id, created_at
+        ))
+
+        position_id = cursor.lastrowid
+
+        # Insert holdings (unchanged from previous position)
+        for symbol, qty in current_position.items():
+            if symbol != "CASH":
+                cursor.execute("""
+                    INSERT INTO holdings (position_id, symbol, quantity)
+                    VALUES (?, ?, ?)
+                """, (position_id, symbol, qty))
+
+        conn.commit()
+
+    except Exception as e:
+        conn.rollback()
+        raise
+    finally:
+        conn.close()
+```
+
+#### 3. System Prompt Builder Updates
+
+**Modified Function (prompts/agent_prompt.py):**
+
+```python
+def get_agent_system_prompt(today_date: str, signature: str) -> str:
+    """Build system prompt with database position queries."""
+    from tools.general_tools import get_config_value
+
+    print(f"signature: {signature}")
+    print(f"today_date: {today_date}")
+
+    # Get job_id from runtime config
+    job_id = get_config_value("JOB_ID")
+    if not job_id:
+        raise ValueError("JOB_ID not found in runtime config")
+
+    # Query database for yesterday's position
+    today_init_position = get_today_init_position_from_db(
+        today_date, signature, job_id
+    )
+
+    # Get prices (unchanged)
+    yesterday_buy_prices, yesterday_sell_prices = get_yesterday_open_and_close_price(
+        today_date, all_nasdaq_100_symbols
+    )
+    today_buy_price = get_open_prices(today_date, all_nasdaq_100_symbols)
+    yesterday_profit = get_yesterday_profit(
+        today_date, yesterday_buy_prices, yesterday_sell_prices, today_init_position
+    )
+
+    return agent_system_prompt.format(
+        date=today_date,
+        positions=today_init_position,
+        STOP_SIGNAL=STOP_SIGNAL,
+        yesterday_close_price=yesterday_sell_prices,
+        today_buy_price=today_buy_price,
+        yesterday_profit=yesterday_profit
+    )
+```
+
+#### 4. No-Trade Handler Updates
+
+**Modified Method (agent/base_agent/base_agent.py):**
+
+```python
+async def _handle_trading_result(self, today_date: str) -> None:
+    """Handle trading results with database writes."""
+    from tools.general_tools import get_config_value
+    from tools.price_tools import add_no_trade_record_to_db
+
+    if_trade = get_config_value("IF_TRADE")
+
+    if if_trade:
+        write_config_value("IF_TRADE", False)
+        print("✅ Trading completed")
+    else:
+        print("📊 No trading, maintaining positions")
+
+        # Get context from runtime config
+        job_id = get_config_value("JOB_ID")
+        session_id = self.context_injector.session_id if self.context_injector else None
+
+        if not job_id or not session_id:
+            raise ValueError("Missing JOB_ID or session_id for no-trade record")
+
+        # Write no-trade record to database
+        add_no_trade_record_to_db(
+            today_date,
+            self.signature,
+            job_id,
+            session_id
+        )
+
+        write_config_value("IF_TRADE", False)
+```
+
+### Data Flow Summary
+
+**Complete Execution Sequence:**
+
+1. `ModelDayExecutor.__init__()`:
+   - Create runtime config file with TODAY_DATE, SIGNATURE, JOB_ID
+
+2. `ModelDayExecutor.execute_async()`:
+   - Create trading_sessions record → get session_id
+   - Initialize starting position (if first day)
+   - Set RUNTIME_ENV_PATH environment variable
+   - Initialize agent (without ContextInjector)
+   - Create ContextInjector(date, model_sig, job_id, session_id)
+   - Call agent.set_context(context_injector)
+   - Run trading session
+
+3. `BaseAgent.run_trading_session()`:
+   - Build system prompt → queries database for yesterday's position
+   - AI agent analyzes and decides
+   - Calls buy/sell tools → ContextInjector injects parameters
+   - Trade tools write to database
+   - If no trade: add_no_trade_record_to_db()
+
+4. Position Query Flow:
+   - System prompt needs yesterday's position
+   - `get_today_init_position_from_db(today_date, signature, job_id)`
+   - Query: `SELECT positions + holdings WHERE job_id=? AND model=? AND date<? ORDER BY date DESC, action_id DESC LIMIT 1`
+   - Reconstruct position dict from results
+   - Return to system prompt builder
+
+### Testing Strategy
+
+**Critical Test Cases:**
+
+1. **First Trading Day**
+   - No previous position in database
+   - Returns `{"CASH": 10000.0}`
+   - System prompt shows available cash
+   - Initial position created with action_id=0
+
+2. **Subsequent Trading Days**
+   - Query finds previous position
+   - System prompt shows yesterday's holdings
+   - Action_id increments properly
+
+3. **No-Trade Days**
+   - Agent outputs `<FINISH_SIGNAL>` without trading
+   - `add_no_trade_record_to_db()` creates record
+   - Holdings unchanged
+   - Portfolio value calculated
+
+4. **ContextInjector Values**
+   - All parameters non-None
+   - Debug log shows correct injection
+   - Trade tools validate successfully
+
+**Edge Cases:**
+
+- Multiple models, same job (different signatures)
+- Date gaps (weekends) - query finds Friday on Monday
+- Mid-simulation restart - resumes from last position
+- Empty holdings (only CASH)
+
+**Validation Points:**
+
+- Log ContextInjector values at injection
+- Log database query results
+- Verify initial position created
+- Check session_id links positions
+
+## Implementation Checklist
+
+### Phase 1: ContextInjector Refactor
+- [ ] Remove ContextInjector creation from BaseAgent.__init__()
+- [ ] Add BaseAgent.set_context() method
+- [ ] Update ModelDayExecutor to create and inject ContextInjector
+- [ ] Add debug logging for injected values
+
+### Phase 2: Database Position Functions
+- [ ] Implement get_today_init_position_from_db()
+- [ ] Implement add_no_trade_record_to_db()
+- [ ] Add database error handling
+- [ ] Add logging for query results
+
+### Phase 3: Integration
+- [ ] Update get_agent_system_prompt() to use database queries
+- [ ] Update _handle_trading_result() to use database writes
+- [ ] Remove/deprecate old file-based functions
+- [ ] Test first trading day scenario
+- [ ] Test subsequent trading days
+- [ ] Test no-trade scenario
+
+### Phase 4: Validation
+- [ ] Run full simulation and verify ContextInjector logs
+- [ ] Verify initial cash appears in system prompt
+- [ ] Verify trades execute successfully
+- [ ] Verify no-trade records created
+- [ ] Check database for correct position records
+
+## Rollback Plan
+
+If issues arise:
+1. Revert ContextInjector changes (keep in __init__)
+2. Temporarily pass correct date via environment variable
+3. Keep file-based functions as fallback
+4. Debug database queries in isolation
+
+## Success Criteria
+
+1. ContextInjector logs show all non-None values
+2. System prompt displays initial $10,000 cash
+3. Trade tools successfully execute buy/sell operations
+4. No FileNotFoundError exceptions
+5. Database contains correct position records
+6. AI agent can complete full trading day
+
+## Notes
+
+- File-based functions marked as deprecated but not removed (backward compatibility)
+- Database queries use deployment_config for automatic prod/dev resolution
+- Initial cash value should eventually be read from config, not hardcoded
+- Consider adding database connection pooling for performance

prompts/agent_prompt.py

@@ -68,14 +68,24 @@ When you think your task is complete, output
 def get_agent_system_prompt(today_date: str, signature: str) -> str:
     print(f"signature: {signature}")
     print(f"today_date: {today_date}")
+
+    # Get job_id from runtime config
+    job_id = get_config_value("JOB_ID")
+    if not job_id:
+        raise ValueError("JOB_ID not found in runtime config")
+
+    # Query database for yesterday's position
+    from tools.price_tools import get_today_init_position_from_db
+    today_init_position = get_today_init_position_from_db(today_date, signature, job_id)
+
     # Get yesterday's buy and sell prices
     yesterday_buy_prices, yesterday_sell_prices = get_yesterday_open_and_close_price(today_date, all_nasdaq_100_symbols)
     today_buy_price = get_open_prices(today_date, all_nasdaq_100_symbols)
-    today_init_position = get_today_init_position(today_date, signature)
     yesterday_profit = get_yesterday_profit(today_date, yesterday_buy_prices, yesterday_sell_prices, today_init_position)
+
     return agent_system_prompt.format(
-        date=today_date, 
-        positions=today_init_position, 
+        date=today_date,
+        positions=today_init_position,
         STOP_SIGNAL=STOP_SIGNAL,
         yesterday_close_price=yesterday_sell_prices,
         today_buy_price=today_buy_price,

tools/price_tools.py

@@ -299,7 +299,172 @@ def add_no_trade_record(today_date: str, modelname: str):
 
     with position_file.open("a", encoding="utf-8") as f:
         f.write(json.dumps(save_item) + "\n")
-    return 
+    return
+
+
+def get_today_init_position_from_db(
+    today_date: str,
+    modelname: str,
+    job_id: str
+) -> Dict[str, float]:
+    """
+    Query yesterday's position from SQLite database.
+
+    Args:
+        today_date: Current trading date (YYYY-MM-DD)
+        modelname: Model signature
+        job_id: Job UUID
+
+    Returns:
+        Position dict: {"AAPL": 50, "MSFT": 30, "CASH": 5000.0}
+        If no position exists: {"CASH": 10000.0} (initial cash)
+    """
+    import logging
+    from api.database import get_db_connection
+
+    logger = logging.getLogger(__name__)
+
+    db_path = "data/jobs.db"
+    conn = get_db_connection(db_path)
+    cursor = conn.cursor()
+
+    try:
+        # Get most recent position before today
+        cursor.execute("""
+            SELECT p.id, p.cash
+            FROM positions p
+            WHERE p.job_id = ? AND p.model = ? AND p.date < ?
+            ORDER BY p.date DESC, p.action_id DESC
+            LIMIT 1
+        """, (job_id, modelname, today_date))
+
+        row = cursor.fetchone()
+
+        if not row:
+            # First day - return initial cash
+            logger.info(f"No previous position found for {modelname}, returning initial cash")
+            return {"CASH": 10000.0}
+
+        position_id, cash = row
+        position_dict = {"CASH": cash}
+
+        # Get holdings for this position
+        cursor.execute("""
+            SELECT symbol, quantity
+            FROM holdings
+            WHERE position_id = ?
+        """, (position_id,))
+
+        for symbol, quantity in cursor.fetchall():
+            position_dict[symbol] = quantity
+
+        logger.debug(f"Loaded position for {modelname}: {position_dict}")
+        return position_dict
+
+    except Exception as e:
+        logger.error(f"Database error in get_today_init_position_from_db: {e}")
+        raise
+    finally:
+        conn.close()
+
+
+def add_no_trade_record_to_db(
+    today_date: str,
+    modelname: str,
+    job_id: str,
+    session_id: int
+) -> None:
+    """
+    Create no-trade position record in SQLite database.
+
+    Args:
+        today_date: Current trading date (YYYY-MM-DD)
+        modelname: Model signature
+        job_id: Job UUID
+        session_id: Trading session ID
+    """
+    import logging
+    from api.database import get_db_connection
+    from agent_tools.tool_trade import get_current_position_from_db
+    from datetime import datetime
+
+    logger = logging.getLogger(__name__)
+
+    db_path = "data/jobs.db"
+    conn = get_db_connection(db_path)
+    cursor = conn.cursor()
+
+    try:
+        # Get current position
+        current_position, next_action_id = get_current_position_from_db(
+            job_id, modelname, today_date
+        )
+
+        # Calculate portfolio value
+        cash = current_position.get("CASH", 0.0)
+        portfolio_value = cash
+
+        # Add stock values
+        for symbol, qty in current_position.items():
+            if symbol != "CASH":
+                try:
+                    price = get_open_prices(today_date, [symbol])[f'{symbol}_price']
+                    portfolio_value += qty * price
+                except KeyError:
+                    logger.warning(f"Price not found for {symbol} on {today_date}")
+                    pass
+
+        # Get previous value for P&L
+        cursor.execute("""
+            SELECT portfolio_value
+            FROM positions
+            WHERE job_id = ? AND model = ? AND date < ?
+            ORDER BY date DESC, action_id DESC
+            LIMIT 1
+        """, (job_id, modelname, today_date))
+
+        row = cursor.fetchone()
+        previous_value = row[0] if row else 10000.0
+
+        daily_profit = portfolio_value - previous_value
+        daily_return_pct = (daily_profit / previous_value * 100) if previous_value > 0 else 0
+
+        # Insert position record
+        created_at = datetime.utcnow().isoformat() + "Z"
+
+        cursor.execute("""
+            INSERT INTO positions (
+                job_id, date, model, action_id, action_type,
+                cash, portfolio_value, daily_profit, daily_return_pct,
+                session_id, created_at
+            )
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        """, (
+            job_id, today_date, modelname, next_action_id, "no_trade",
+            cash, portfolio_value, daily_profit, daily_return_pct,
+            session_id, created_at
+        ))
+
+        position_id = cursor.lastrowid
+
+        # Insert holdings (unchanged from previous position)
+        for symbol, qty in current_position.items():
+            if symbol != "CASH":
+                cursor.execute("""
+                    INSERT INTO holdings (position_id, symbol, quantity)
+                    VALUES (?, ?, ?)
+                """, (position_id, symbol, qty))
+
+        conn.commit()
+        logger.info(f"Created no-trade record for {modelname} on {today_date}")
+
+    except Exception as e:
+        conn.rollback()
+        logger.error(f"Database error in add_no_trade_record_to_db: {e}")
+        raise
+    finally:
+        conn.close()
+
 
 if __name__ == "__main__":
     today_date = get_config_value("TODAY_DATE")