chore(master): release 1.3.11 (#45 )

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
fix(ci): lazy import logger in predix.py and cli.py to avoid ImportError in test env
2026-07-29 16:37:43 +00:00 · 2026-05-01 13:58:55 +02:00 · 2026-05-01 13:58:16 +02:00 · 2026-05-01 13:50:38 +02:00 · 2026-05-01 13:49:58 +02:00 · 2026-05-01 13:43:52 +02:00
77 changed files with 4156 additions and 381 deletions
@@ -12,7 +12,7 @@ jobs:
  release-please:
    runs-on: ubuntu-latest
    steps:
-      - uses: googleapis/release-please-action@v4
+      - uses: googleapis/release-please-action@v5
        with:
          token: ${{ secrets.GITHUB_TOKEN }}
          config-file: release-please-config.json
@@ -1,3 +1,3 @@
 {
-  ".": "1.3.0"
+  ".": "1.3.11"
 }
@@ -1,5 +1,125 @@
 # Changelog

+## [1.3.11](https://github.com/TPTBusiness/Predix/compare/v1.3.10...v1.3.11) (2026-05-01)
+
+
+### Bug Fixes
+
+* **ci:** lazy import logger in predix.py and cli.py to avoid ImportError in test env ([60763e8](https://github.com/TPTBusiness/Predix/commit/60763e8eae34f41865ba8e5e65bdfde13b564b4b))
+
+## [1.3.10](https://github.com/TPTBusiness/Predix/compare/v1.3.9...v1.3.10) (2026-05-01)
+
+
+### Bug Fixes
+
+* **security:** replace remaining assert statements with proper error handling ([928533d](https://github.com/TPTBusiness/Predix/commit/928533d9a81bd5062f07458fbf94d3c7fe347775))
+
+## [1.3.9](https://github.com/TPTBusiness/Predix/compare/v1.3.8...v1.3.9) (2026-05-01)
+
+
+### Bug Fixes
+
+* **security:** resolve path-injection, B701, B101, B112 Bandit alerts ([20b89a0](https://github.com/TPTBusiness/Predix/commit/20b89a061843b39836e975f158404e8e2d4627cd))
+
+## [1.3.8](https://github.com/TPTBusiness/Predix/compare/v1.3.7...v1.3.8) (2026-04-30)
+
+
+### Bug Fixes
+
+* **deps:** relax aiohttp constraint to &gt;=3.13.4 for litellm compatibility ([34ab192](https://github.com/TPTBusiness/Predix/commit/34ab1923a887089eb36e5cbad6cb8df16f0333ca))
+* **qlib:** correct indentation in except blocks in quant_proposal and factor_runner ([8143451](https://github.com/TPTBusiness/Predix/commit/8143451e8c0ead01c4d86d19669268c7bfb15fac))
+* **security:** replace eval() with ast.literal_eval in finetune validator (B307) ([0508caf](https://github.com/TPTBusiness/Predix/commit/0508caf9140d210b823fefefa28ee535ec85a0ae))
+* **security:** replace shell=True subprocess calls with list args in env.py (B602) ([2012d5a](https://github.com/TPTBusiness/Predix/commit/2012d5ae4e77cc2f1ab9a48beaaac5a74695d083))
+* **security:** resolve path-injection and add nosec for safe temp paths (B108, py/path-injection) ([6727480](https://github.com/TPTBusiness/Predix/commit/67274803bd1d14e5d1df9a063f46b2edb8501a2b))
+
+## [1.3.7](https://github.com/TPTBusiness/Predix/compare/v1.3.6...v1.3.7) (2026-04-30)
+
+
+### Bug Fixes
+
+* **security:** nosec for B608/B701 false positives in UI and template code ([5eb5d7e](https://github.com/TPTBusiness/Predix/commit/5eb5d7e8fdbe90e0dced83fef4e09f5a33e96b2b))
+* **security:** replace eval() with ast.literal_eval and add request timeouts (B307, B113) ([3301ada](https://github.com/TPTBusiness/Predix/commit/3301ada697ca7d3afa1a188d2a76a87ae98b4529))
+* **security:** replace shell=True subprocess calls with list args (B602) ([13c08f4](https://github.com/TPTBusiness/Predix/commit/13c08f4ce6813eb7c314087921ec8c0f40074bd7))
+
+## [1.3.6](https://github.com/TPTBusiness/Predix/compare/v1.3.5...v1.3.6) (2026-04-30)
+
+
+### Bug Fixes
+
+* **security:** real fix for B110 (logging in factor_proposal.py [#746](https://github.com/TPTBusiness/Predix/issues/746)) ([16624e0](https://github.com/TPTBusiness/Predix/commit/16624e0bd966ae4d24c4a3eb42bbc31c11da3136))
+* **security:** real fix for B110 (logging in factor_runner.py [#744](https://github.com/TPTBusiness/Predix/issues/744)) ([88cf0fb](https://github.com/TPTBusiness/Predix/commit/88cf0fb8828b11c97f2f3ae2881a4900b020c6f0))
+* **security:** real fix for B110 (logging in quant_proposal.py [#741](https://github.com/TPTBusiness/Predix/issues/741)) ([7cf2a64](https://github.com/TPTBusiness/Predix/commit/7cf2a644f553b054bd4b0607ea51e5372e68d90a))
+* **security:** real fix for B110 (logging in quant_proposal.py [#741](https://github.com/TPTBusiness/Predix/issues/741)) ([ef985f8](https://github.com/TPTBusiness/Predix/commit/ef985f86035d8dca707c60137e6508349a0c4ae6))
+* **security:** real fix for B404/B603 (sys.executable in factor_runner.py [#745](https://github.com/TPTBusiness/Predix/issues/745)) ([819655a](https://github.com/TPTBusiness/Predix/commit/819655aaa3efa76596d60501d0e8ca365df3e5e2))
+* **security:** revert broken read_pickle encoding arg in kaggle template (B301) ([3574907](https://github.com/TPTBusiness/Predix/commit/35749073c91e69f63ddaad61dae3f2b799327e63))
+* **security:** validate SQL identifiers in _add_column_if_not_exists (B608) ([e10dfa2](https://github.com/TPTBusiness/Predix/commit/e10dfa2576038e911f83595d3b466c261bc0cd54))
+* **security:** whitelist-validate metric column in get_top_factors (B608) ([e50519f](https://github.com/TPTBusiness/Predix/commit/e50519fe066e68aec2f19b83df4f643c3c22053d))
+
+## [1.3.5](https://github.com/TPTBusiness/Predix/compare/v1.3.4...v1.3.5) (2026-04-27)
+
+
+### Bug Fixes
+
+* **auto-fixer:** add five new factor code fixes for groupby/apply errors ([449c8fd](https://github.com/TPTBusiness/Predix/commit/449c8fd70a327e604dcca122e4a134f0cca918e4))
+* **auto-fixer:** add four new factor code fixes for common runtime errors ([40484f6](https://github.com/TPTBusiness/Predix/commit/40484f6d300425da481f1edd325da4acbc06ec7d))
+* **auto-fixer:** add groupby([level=N,'date']) SyntaxError fix ([ca77c00](https://github.com/TPTBusiness/Predix/commit/ca77c005bea4abdd8854c1de2b0e8d03b7742161))
+* **auto-fixer:** disable _fix_min_periods for intraday data ([77b0740](https://github.com/TPTBusiness/Predix/commit/77b0740f059349df7e769a378af728aa33b2070e))
+* **auto-fixer:** fix chained groupby(level=N).groupby('date') pattern ([7d5fe32](https://github.com/TPTBusiness/Predix/commit/7d5fe32b31a19ce8b04bd8f5a430720fdb748f7a))
+* **auto-fixer:** fix df.loc[instrument] DateParseError on MultiIndex frames ([b7860ea](https://github.com/TPTBusiness/Predix/commit/b7860eafc0ad26384947ce0510ecf4e9f3425807))
+* **auto-fixer:** fix df['instrument'] KeyError on MultiIndex frames ([aad6bd1](https://github.com/TPTBusiness/Predix/commit/aad6bd1c7c720b3d486e0cf248337f32394773b1))
+* **auto-fixer:** fix two assignment-target bugs in instrument column fixers ([421eedf](https://github.com/TPTBusiness/Predix/commit/421eedffed4b883c24397dc5581c019a3985277f))
+* **auto-fixer:** preserve date dimension in groupby(['instrument','date']) fix ([b58fdd8](https://github.com/TPTBusiness/Predix/commit/b58fdd8be43720b5d4363e0f8de9a01591d4d2dc))
+* **auto-fixer:** remove ddof from rolling() args, not only from std()/var() ([b0fc328](https://github.com/TPTBusiness/Predix/commit/b0fc328d0d4a041c65d8eeb32cb3f2bb86568406))
+* **auto-fixer:** strip spurious .reset_index() after .transform() calls ([8708aae](https://github.com/TPTBusiness/Predix/commit/8708aae6e08728cda1875c775a76dc92e43576f3))
+* **loop:** prevent step_idx advance on unhandled exceptions + fix consecutive assistant messages ([5ec4ad1](https://github.com/TPTBusiness/Predix/commit/5ec4ad1b96b5b99ef42bea7bb828cb1ef709a688))
+
+## [1.3.4](https://github.com/TPTBusiness/Predix/compare/v1.3.3...v1.3.4) (2026-04-27)
+
+
+### Bug Fixes
+
+* **auto-fixer:** add five new factor code fixes for groupby/apply errors ([449c8fd](https://github.com/TPTBusiness/Predix/commit/449c8fd70a327e604dcca122e4a134f0cca918e4))
+* **auto-fixer:** add four new factor code fixes for common runtime errors ([40484f6](https://github.com/TPTBusiness/Predix/commit/40484f6d300425da481f1edd325da4acbc06ec7d))
+* **auto-fixer:** add groupby([level=N,'date']) SyntaxError fix ([ca77c00](https://github.com/TPTBusiness/Predix/commit/ca77c005bea4abdd8854c1de2b0e8d03b7742161))
+* **auto-fixer:** disable _fix_min_periods for intraday data ([77b0740](https://github.com/TPTBusiness/Predix/commit/77b0740f059349df7e769a378af728aa33b2070e))
+* **auto-fixer:** fix chained groupby(level=N).groupby('date') pattern ([7d5fe32](https://github.com/TPTBusiness/Predix/commit/7d5fe32b31a19ce8b04bd8f5a430720fdb748f7a))
+* **auto-fixer:** fix df.loc[instrument] DateParseError on MultiIndex frames ([b7860ea](https://github.com/TPTBusiness/Predix/commit/b7860eafc0ad26384947ce0510ecf4e9f3425807))
+* **auto-fixer:** fix df['instrument'] KeyError on MultiIndex frames ([aad6bd1](https://github.com/TPTBusiness/Predix/commit/aad6bd1c7c720b3d486e0cf248337f32394773b1))
+* **auto-fixer:** preserve date dimension in groupby(['instrument','date']) fix ([b58fdd8](https://github.com/TPTBusiness/Predix/commit/b58fdd8be43720b5d4363e0f8de9a01591d4d2dc))
+* **auto-fixer:** remove ddof from rolling() args, not only from std()/var() ([b0fc328](https://github.com/TPTBusiness/Predix/commit/b0fc328d0d4a041c65d8eeb32cb3f2bb86568406))
+* **backtest:** replace broken MC permutation test with binomial win-rate test ([c38d894](https://github.com/TPTBusiness/Predix/commit/c38d89478f586825bfca5715a96ca70ccd8791a3))
+* **factors:** detect and correct look-ahead bias in daily-constant factors ([eb490a4](https://github.com/TPTBusiness/Predix/commit/eb490a461b66cbd815ae53ac5205115754712432))
+* **factors:** extend look-ahead rules to session factors and add intraday-factor guidance ([c24c100](https://github.com/TPTBusiness/Predix/commit/c24c100442d6487686c0578de0b32d240fcbf215))
+* **loop:** compress old experiment history in proposal prompt to reduce context size ([4bf90a9](https://github.com/TPTBusiness/Predix/commit/4bf90a905ba8b2aba2a818191c19998088cccaaf))
+* **loop:** prevent step_idx advance on unhandled exceptions + fix consecutive assistant messages ([5ec4ad1](https://github.com/TPTBusiness/Predix/commit/5ec4ad1b96b5b99ef42bea7bb828cb1ef709a688))
+
+## [1.3.3](https://github.com/TPTBusiness/Predix/compare/v1.3.2...v1.3.3) (2026-04-25)
+
+
+### Bug Fixes
+
+* **backtest:** replace broken MC permutation test with binomial win-rate test ([c38d894](https://github.com/TPTBusiness/Predix/commit/c38d89478f586825bfca5715a96ca70ccd8791a3))
+* **factors:** detect and correct look-ahead bias in daily-constant factors ([eb490a4](https://github.com/TPTBusiness/Predix/commit/eb490a461b66cbd815ae53ac5205115754712432))
+* **factors:** extend look-ahead rules to session factors and add intraday-factor guidance ([c24c100](https://github.com/TPTBusiness/Predix/commit/c24c100442d6487686c0578de0b32d240fcbf215))
+* **loop:** compress old experiment history in proposal prompt to reduce context size ([4bf90a9](https://github.com/TPTBusiness/Predix/commit/4bf90a905ba8b2aba2a818191c19998088cccaaf))
+* **strategies:** guard against None IC in acceptance check, disable slow wf_rolling ([2197f52](https://github.com/TPTBusiness/Predix/commit/2197f52150a50ef38d9e70991d7e48c8c30caec4))
+* **strategies:** handle None ic/sharpe/dd in rejected strategy log output ([ad2ad3a](https://github.com/TPTBusiness/Predix/commit/ad2ad3ab3360ea75ed3bbc90c12098b9c5cc0114))
+
+## [1.3.2](https://github.com/TPTBusiness/Predix/compare/v1.3.1...v1.3.2) (2026-04-23)
+
+
+### Bug Fixes
+
+* **strategies:** guard against None IC in acceptance check, disable slow wf_rolling ([2197f52](https://github.com/TPTBusiness/Predix/commit/2197f52150a50ef38d9e70991d7e48c8c30caec4))
+* **strategies:** handle None ic/sharpe/dd in rejected strategy log output ([ad2ad3a](https://github.com/TPTBusiness/Predix/commit/ad2ad3ab3360ea75ed3bbc90c12098b9c5cc0114))
+
+## [1.3.1](https://github.com/TPTBusiness/Predix/compare/v1.3.0...v1.3.1) (2026-04-21)
+
+
+### Bug Fixes
+
+* **deps:** bump python-dotenv to &gt;=1.2.2 (CVE symlink overwrite) ([126ae7d](https://github.com/TPTBusiness/Predix/commit/126ae7d5fb556b677d09d10221862a0d648d697a))
+
 ## [1.3.0](https://github.com/TPTBusiness/Predix/compare/v1.2.2...v1.3.0) (2026-04-21)


@@ -18,6 +18,13 @@ load_dotenv(Path(__file__).parent / ".env")
 import typer
 from rich.console import Console

+try:
+    from rdagent.utils.env import logger
+except ImportError:
+    import logging
+
+    logger = logging.getLogger(__name__)
+
 app = typer.Typer(help="Predix - AI Quantitative Trading Agent")
 console = Console()

@@ -510,6 +517,7 @@ def top(
            if data.get("status") == "success" and data.get("ic") is not None:
                results.append(data)
        except Exception:
+            logger.warning("Failed to load factor file %s", f, exc_info=True)
            continue

    if not results:
@@ -659,6 +667,7 @@ def portfolio(
            if data.get("status") == "success" and data.get("ic") is not None:
                results.append(data)
        except Exception:
+            logger.warning("Failed to load factor file %s", f, exc_info=True)
            continue

    if not results:
@@ -956,6 +965,7 @@ def portfolio_simple(
            if data.get("status") == "success" and data.get("ic") is not None:
                results.append(data)
        except Exception:
+            logger.warning("Failed to load factor file %s", f, exc_info=True)
            continue

    if not results:
@@ -1337,6 +1347,7 @@ def build_strategies_ai(
            if data.get("status") == "success" and data.get("ic") is not None:
                factors.append(data)
        except Exception:
+            logger.warning("Failed to load factor file %s", f, exc_info=True)
            continue

    if len(factors) < 10:
@@ -1552,6 +1563,7 @@ def _load_strategies():
            try:
                raw = json.loads(p.read_text())
            except Exception:
+                logger.warning("Failed to load strategy file %s", p, exc_info=True)
                continue
            if not isinstance(raw, dict):
                continue
@@ -27,6 +27,13 @@ import typer
 from rich.console import Console
 from typing_extensions import Annotated

+try:
+    from rdagent.utils.env import logger
+except ImportError:
+    import logging
+
+    logger = logging.getLogger(__name__)
+
 from rdagent.app.data_science.loop import main as data_science
 from rdagent.app.finetune.llm.loop import main as llm_finetune
 from rdagent.app.general_model.general_model import (
@@ -882,6 +889,7 @@ def optimize_portfolio_cli(
                if data.get("status") == "accepted":
                    strategies.append(data)
            except Exception:
+                logger.warning("Failed to load strategy file %s", f, exc_info=True)
                continue

        if not strategies:
@@ -1251,7 +1259,7 @@ def start_loop_cli(
    script_dir = str(Path(__file__).parent.parent.parent.parent)
    generator = f"python {script_dir}/scripts/predix_smart_strategy_gen.py"
    logfile = f"{script_dir}/results/logs/generator_loop.log"
-    pidfile = "/tmp/predix_loop.pid"
+    pidfile = "/tmp/predix_loop.pid"  # nosec B108 — administrative PID file, single-process daemon

    os.makedirs(f"{script_dir}/results/logs", exist_ok=True)

@@ -201,6 +201,5 @@ class DataScienceBasePropSetting(KaggleBasePropSetting):
 DS_RD_SETTING = DataScienceBasePropSetting()

 # enable_cross_trace_diversity and llm_select_hypothesis should not be true at the same time
-assert not (
-    DS_RD_SETTING.enable_cross_trace_diversity and DS_RD_SETTING.llm_select_hypothesis
-), "enable_cross_trace_diversity and llm_select_hypothesis cannot be true at the same time"
+if DS_RD_SETTING.enable_cross_trace_diversity and DS_RD_SETTING.llm_select_hypothesis:
+    raise ValueError("enable_cross_trace_diversity and llm_select_hypothesis cannot be true at the same time")
@@ -58,18 +58,18 @@ def main(

    if user_target_scenario:
        FT_RD_SETTING.user_target_scenario = user_target_scenario
-    assert (
-        FT_RD_SETTING.user_target_scenario is None
-    ), "user_target_scenario is not yet supported, please specify via benchmark and benchmark_description"
+    if FT_RD_SETTING.user_target_scenario is not None:
+        raise ValueError("user_target_scenario is not yet supported, please specify via benchmark and benchmark_description")
    if upper_data_size_limit:
        FT_RD_SETTING.upper_data_size_limit = upper_data_size_limit
        logger.info(f"Set upper_data_size_limit to {FT_RD_SETTING.upper_data_size_limit}")
    if benchmark and benchmark_description:
        FT_RD_SETTING.target_benchmark = benchmark
        FT_RD_SETTING.benchmark_description = benchmark_description
-    assert FT_RD_SETTING.user_target_scenario or (
-        FT_RD_SETTING.target_benchmark and FT_RD_SETTING.benchmark_description
-    ), "Either user_target_scenario or target_benchmark must be specified for LLM fine-tuning."
+    if not (
+        FT_RD_SETTING.user_target_scenario or (FT_RD_SETTING.target_benchmark and FT_RD_SETTING.benchmark_description)
+    ):
+        raise ValueError("Either user_target_scenario or target_benchmark must be specified for LLM fine-tuning.")

    # Update configuration with provided parameters
    if dataset:
@@ -82,9 +82,8 @@ def main(
    model_target = FT_RD_SETTING.base_model if FT_RD_SETTING.base_model else "auto selected model"

    # Temporary assertion until auto-selection is implemented
-    assert (
-        FT_RD_SETTING.base_model is not None
-    ), "Base model auto selection not yet supported, please specify via --base-model"
+    if FT_RD_SETTING.base_model is None:
+        raise ValueError("Base model auto selection not yet supported, please specify via --base-model")

    logger.info(f"Starting LLM fine-tuning on dataset='{data_set_target}' with model='{model_target}'")

@@ -24,46 +24,12 @@ from rdagent.app.finetune.llm.ui.ft_summary import render_job_summary

 DEFAULT_LOG_BASE = "log/"

+from rdagent.core.utils import safe_resolve_path
+

 def validate_path_within_cwd(user_path: Path) -> Path:
-    """
-    Validate that a user-provided path is within the current working directory.
-
-    Security: This function prevents path traversal attacks by:
-    1. Resolving the path to its absolute canonical form
-    2. Verifying it's within the CWD boundary using a normalized common prefix
-    3. Rejecting paths outside the boundary with ValueError
-
-    Parameters
-    ----------
-    user_path : Path
-        User-provided path to validate
-
-    Returns
-    -------
-    Path
-        Resolved absolute path if valid
-
-    Raises
-    ------
-    ValueError
-        If path is outside the current working directory
-    """
    safe_root = Path.cwd().resolve()
-    # Expand any user home reference and resolve without requiring the path to exist.
-    resolved_path = user_path.expanduser().resolve(strict=False)
-
-    # Ensure the resolved path is absolute and remains within the safe root.
-    safe_root_str = str(safe_root)
-    resolved_str = str(resolved_path)
-    common = os.path.commonpath([safe_root_str, resolved_str])
-    if common != safe_root_str:
-        raise ValueError("Path is outside the allowed project directory")
-
-    # This will raise ValueError if resolved_path is not within safe_root
-    resolved_path.relative_to(safe_root)
-
-    return resolved_path
+    return safe_resolve_path(user_path, safe_root)


 def get_job_options(base_path: Path, safe_root: Path | None = None) -> list[str]:
@@ -141,19 +107,14 @@ def main():
            st.header("Job")
            base_folder = st.text_input("Base Folder", value=default_log, key="base_folder_input")

-            # Normalize and validate the base folder against the configured log root
-            root_real = os.path.realpath(str(Path(default_log).expanduser()))
-            folder_real = os.path.realpath(str(Path(base_folder).expanduser()))
-            if folder_real == root_real or folder_real.startswith(root_real + os.sep):
-                base_path = Path(folder_real)
-                safe_root = Path(root_real)
-            else:
+            safe_root = Path(default_log).expanduser().resolve()
+            try:
+                base_path = safe_resolve_path(Path(base_folder), safe_root)
+            except ValueError:
                st.error("Invalid base folder: must be within the configured log directory.")
-                safe_root = Path(root_real)
                base_path = safe_root

-            # base_path is validated against safe_root – nosec B614
-            job_options = get_job_options(base_path, safe_root)  # nosec B614 – validated above
+            job_options = get_job_options(base_path, safe_root)
            if job_options:
                selected_job = st.selectbox("Select Job", job_options, key="job_select")
                if selected_job.startswith("."):
@@ -13,6 +13,7 @@ from typing import Any
 import streamlit as st

 from rdagent.app.finetune.llm.ui.config import EVALUATOR_CONFIG, EventType
+from rdagent.core.utils import safe_resolve_path
 from rdagent.log.storage import FileStorage


@@ -89,11 +90,10 @@ def extract_stage(tag: str) -> str:
 def get_valid_sessions(log_folder: Path, safe_root: Path | None = None) -> list[str]:
    """Get list of valid session directories, optionally validating against a safe root."""
    if safe_root is not None:
-        root_real = os.path.realpath(str(safe_root.expanduser()))
-        folder_real = os.path.realpath(str(log_folder.expanduser()))
-        if not (folder_real == root_real or folder_real.startswith(root_real + os.sep)):
+        try:
+            log_folder = safe_resolve_path(log_folder, safe_root)
+        except ValueError:
            return []
-        log_folder = Path(folder_real)

    if not log_folder.exists():
        return []
@@ -373,13 +373,11 @@ def parse_event(tag: str, content: Any, timestamp: datetime) -> Event | None:
@st.cache_data(ttl=300, hash_funcs={Path: str})
 def load_ft_session(log_path: Path, safe_root: Path | None = None) -> Session:
    """Load events into hierarchical session structure, optionally validating against safe root."""
-    # Validate path is within safe_root if provided
    if safe_root is not None:
-        root_real = os.path.realpath(str(safe_root.expanduser()))
-        path_real = os.path.realpath(str(log_path.expanduser()))
-        if not (path_real == root_real or path_real.startswith(root_real + os.sep)):
+        try:
+            log_path = safe_resolve_path(log_path, safe_root)
+        except ValueError:
            return Session()
-        log_path = Path(path_real)

    session = Session()
    storage = FileStorage(log_path)
@@ -78,7 +78,8 @@ class QuantRDLoop(RDLoop):
        while True:
            if self.get_unfinished_loop_cnt(self.loop_idx) < RD_AGENT_SETTINGS.get_max_parallel():
                hypo = self._propose()
-                assert hypo.action in ["factor", "model"]
+                if hypo.action not in ["factor", "model"]:
+                    raise ValueError(f"hypo.action must be 'factor' or 'model', got {hypo.action!r}")
                if hypo.action == "factor":
                    exp = self.factor_hypothesis2experiment.convert(hypo, self.trace)
                else:
@@ -322,6 +323,7 @@ class QuantRDLoop(RDLoop):
                    if data.get("status") == "success" and data.get("ic") is not None:
                        factors.append(data)
                except Exception:
+                    logger.warning("Failed to load factor file %s", f, exc_info=True)
                    continue

            if len(factors) < 10:
@@ -16,55 +16,26 @@ from rdagent.app.rl.ui.components import render_session, render_summary
 from rdagent.app.rl.ui.config import ALWAYS_VISIBLE_TYPES, OPTIONAL_TYPES
 from rdagent.app.rl.ui.data_loader import get_summary, get_valid_sessions, load_session
 from rdagent.app.rl.ui.rl_summary import render_job_summary
+from rdagent.core.utils import safe_resolve_path

 DEFAULT_LOG_BASE = "log/"


 def _safe_resolve(user_input: str | None, safe_root: Path) -> Path:
-    """
-    Resolve user path relative to safe_root; raise ValueError if it escapes.
-
-    Security: This function prevents path traversal attacks by:
-    1. Rejecting null bytes in user input
-    2. Rejecting Windows drive letters (C:\, D:\, etc.)
-    3. Rejecting absolute paths
-    4. Normalizing path to remove .. traversal attempts
-    5. Validating resolved path is within safe_root using a realpath-based check
-
-    All user-provided paths are validated before filesystem access.
-    """
-    # Treat the provided safe_root as trusted and canonicalize it once.
    safe_root = safe_root.expanduser().resolve()
-
-    # Empty input maps to the safe root directory.
    if not user_input:
        return safe_root
-
-    # Security check 1: Reject null bytes (path truncation attack)
    if "\x00" in user_input:
        raise ValueError("Invalid path: contains null byte")
-
    try:
-        # Security check 2: Normalize path to resolve .. and . components
        normalized = os.path.normpath(user_input.strip())
-
-        # Security check 3: Reject Windows drive letters (C:\, D:\, etc.)
        drive, _ = os.path.splitdrive(normalized)
        if drive:
            raise ValueError("Absolute paths with drive letters are not allowed")
-
-        # Security check 4: Reject absolute paths (/, //server/share, etc.)
        if os.path.isabs(normalized):
            raise ValueError("Absolute paths are not allowed")
-
-        # Security check 5: Build candidate path under safe_root and fully resolve it.
-        joined = os.path.join(str(safe_root), normalized)
-        resolved_candidate = os.path.realpath(joined)
-
-        # Security check 6: Validate candidate is within safe_root (prevent path traversal)
-        candidate_path = Path(resolved_candidate)
-        # Reconstruct from trusted safe_root so the returned path is root-derived.
-        return safe_root / candidate_path.relative_to(safe_root)
+        joined = safe_root / normalized
+        return safe_resolve_path(joined, safe_root)
    except (OSError, ValueError) as exc:
        raise ValueError(f"Invalid path outside of allowed root: {user_input}") from exc

@@ -82,7 +53,7 @@ def get_job_options(base_path: Path, safe_root: Path | None = None) -> list[str]

    # Security fix: Validate base_path to prevent path traversal
    try:
-        base_path_resolved = base_path.expanduser().resolve()
+        base_path_resolved = base_path.expanduser().resolve()  # nosec B614 — validated against safe_root below via relative_to()

        if safe_root is not None:
            safe_root_resolved = safe_root.expanduser().resolve()
@@ -203,8 +174,7 @@ def main():
        except ValueError as e:
            st.warning(str(e))
            return
-        # job_path is validated by _safe_resolve() above
-        if job_path.exists():  # nosec B614 – path validated by _safe_resolve
+        if job_path.exists():
            render_job_summary(job_path, safe_root, is_root=is_root_job)
        else:
            st.warning(f"Job folder not found: {job_folder}")
@@ -15,6 +15,7 @@ from typing import Any
 import streamlit as st

 from rdagent.app.rl.ui.config import EventType
+from rdagent.core.utils import safe_resolve_path
 from rdagent.log.storage import FileStorage


@@ -76,11 +77,10 @@ def extract_stage(tag: str) -> str:
 def get_valid_sessions(log_folder: Path, safe_root: Path | None = None) -> list[str]:
    """Get list of valid session directories, optionally validating against a safe root."""
    if safe_root is not None:
-        root_real = os.path.realpath(str(safe_root.expanduser()))
-        folder_real = os.path.realpath(str(log_folder.expanduser()))
-        if not (folder_real == root_real or folder_real.startswith(root_real + os.sep)):
+        try:
+            log_folder = safe_resolve_path(log_folder, safe_root)
+        except ValueError:
            return []
-        log_folder = Path(folder_real)

    if not log_folder.exists():
        return []
@@ -245,13 +245,11 @@ def parse_event(tag: str, content: Any, timestamp: datetime) -> Event | None:
@st.cache_data(ttl=300, hash_funcs={Path: str})
 def load_session(log_path: Path, safe_root: Path | None = None) -> Session:
    """Load events into hierarchical session structure, optionally validating against safe root."""
-    # Validate path is within safe_root if provided
    if safe_root is not None:
-        root_real = os.path.realpath(str(safe_root.expanduser()))
-        path_real = os.path.realpath(str(log_path.expanduser()))
-        if not (path_real == root_real or path_real.startswith(root_real + os.sep)):
+        try:
+            log_path = safe_resolve_path(log_path, safe_root)
+        except ValueError:
            return Session()
-        log_path = Path(path_real)

    session = Session()

@@ -9,6 +9,8 @@ from pathlib import Path
 import pandas as pd
 import streamlit as st

+from rdagent.core.utils import safe_resolve_path
+

 def is_valid_task(task_path: Path) -> bool:
    """Check if directory is a valid RL task (has __session__ subdirectory)"""
@@ -62,14 +64,10 @@ def get_loop_status(task_path: Path, loop_id: int) -> tuple[str, bool | None]:


 def _validate_job_path(job_path: Path, safe_root: Path) -> Path:
-    """Resolve and validate that job_path stays within safe_root."""
-    resolved_root = safe_root.expanduser().resolve()
-    resolved_job = job_path.expanduser().resolve()
    try:
-        # Reconstruct from trusted root so the returned path is root-derived.
-        return resolved_root / resolved_job.relative_to(resolved_root)
+        return safe_resolve_path(job_path, safe_root)
    except ValueError:
-        raise ValueError(f"Job path is outside allowed root {resolved_root}")
+        raise ValueError(f"Job path is outside allowed root {safe_root}")


 def get_max_loops(job_path: Path, safe_root: Path | None = None) -> int:
@@ -54,11 +54,11 @@ def rdagent_info():
    current_version = importlib.metadata.version("rdagent")
    logger.info(f"RD-Agent version: {current_version}")
    api_url = f"https://api.github.com/repos/microsoft/RD-Agent/contents/requirements.txt?ref=main"
-    response = requests.get(api_url)
+    response = requests.get(api_url, timeout=30)
    if response.status_code == 200:
        files = response.json()
        file_url = files["download_url"]
-        file_response = requests.get(file_url)
+        file_response = requests.get(file_url, timeout=30)
        if file_response.status_code == 200:
            all_file_contents = file_response.text.split("\n")
        else:
@@ -71,6 +71,9 @@ class ResultsDatabase:

        self.conn.commit()

+    _ALLOWED_TABLES = frozenset({"factors", "backtest_runs", "loop_results"})
+    _ALLOWED_COL_TYPES = frozenset({"REAL", "TEXT", "INTEGER", "BLOB"})
+
    def _add_column_if_not_exists(self, table: str, column: str, col_type: str) -> None:
        """
        Add a column to a table if it doesn't already exist.
@@ -78,20 +81,24 @@ class ResultsDatabase:
        Parameters
        ----------
        table : str
-            Table name
+            Table name (must be in _ALLOWED_TABLES)
        column : str
-            Column name to add
+            Column name to add (alphanumeric + underscore only)
        col_type : str
-            SQL column type (e.g., 'REAL', 'TEXT')
+            SQL column type (must be in _ALLOWED_COL_TYPES)
        """
+        if table not in self._ALLOWED_TABLES:
+            raise ValueError(f"Unknown table: {table!r}")
+        if not column.replace("_", "").isalnum():
+            raise ValueError(f"Invalid column name: {column!r}")
+        if col_type not in self._ALLOWED_COL_TYPES:
+            raise ValueError(f"Invalid column type: {col_type!r}")
+
        c = self.conn.cursor()
-        try:
-            # Try to query the column - if it fails, it doesn't exist
-            # nosec B608: Internal schema migration, column names are controlled
-            c.execute(f"SELECT {column} FROM {table} LIMIT 1")  # nosec B608
-        except sqlite3.OperationalError:
-            # Column doesn't exist, add it
-            c.execute(f"ALTER TABLE {table} ADD COLUMN {column} {col_type}")  # nosec B608
+        c.execute("SELECT name FROM pragma_table_info(?)", (table,))
+        existing = {row[0] for row in c.fetchall()}
+        if column not in existing:
+            c.execute(f"ALTER TABLE {table} ADD COLUMN {column} {col_type}")
    
    def add_factor(self, name: str, type: str = "unknown") -> int:
        c = self.conn.cursor()
@@ -183,16 +190,18 @@ class ResultsDatabase:
        pd.DataFrame
            DataFrame with factor names and metrics
        """
-        # Map shorthand to full column name
+        _ALLOWED_METRICS = frozenset({
+            'sharpe', 'ic', 'annual_return', 'max_drawdown',
+            'win_rate', 'information_ratio', 'volatility',
+        })
        metric_map = {
-            'sharpe': 'sharpe',
-            'ic': 'ic',
-            'return': 'annual_return',
-            'drawdown': 'max_drawdown',
-            'win_rate': 'win_rate',
+            'sharpe': 'sharpe', 'ic': 'ic', 'return': 'annual_return',
+            'drawdown': 'max_drawdown', 'win_rate': 'win_rate',
            'information_ratio': 'information_ratio',
        }
        col = metric_map.get(metric, metric)
+        if col not in _ALLOWED_METRICS:
+            raise ValueError(f"Unknown metric: {metric!r}")

        return pd.read_sql_query(
            f"""SELECT factor_name, ic, sharpe, annual_return, max_drawdown,
@@ -201,7 +210,7 @@ class ResultsDatabase:
                JOIN factors ON factor_id = factors.id
                WHERE {col} IS NOT NULL
                ORDER BY {col} DESC
-                LIMIT ?""",
+                LIMIT ?""",  # nosec B608 — col is validated against _ALLOWED_METRICS above
            self.conn,
            params=[limit]
        )
@@ -356,11 +356,12 @@ def monte_carlo_trade_pvalue(
    """
    Monte Carlo permutation test on trade-level P&L.

-    Shuffles the order of trade returns ``n_permutations`` times and computes
-    the fraction of runs whose total return is >= the real total return.
+    Runs a one-sided binomial test on trade-level win rate.

-    p < 0.05 → strategy has a statistically significant edge (real return
-    beats 95% of random sequences with the same set of trades).
+    Tests H0: win_rate = 0.5 (random trading) against H1: win_rate > 0.5.
+    The ``n_permutations`` parameter is kept for API compatibility but is unused.
+
+    p < 0.05 → win rate is significantly above 50%, indicating a genuine per-trade edge.

    Parameters
    ----------
@@ -379,14 +380,14 @@ def monte_carlo_trade_pvalue(
    if len(trade_pnl) < 2:
        return 1.0
    trades = trade_pnl.values.copy()
-    real_total = float(trades.sum())
-    rng = np.random.default_rng(seed)
-    beat = 0
-    for _ in range(n_permutations):
-        perm = rng.permutation(trades)
-        if perm.sum() >= real_total:
-            beat += 1
-    return beat / n_permutations
+    # Binomial test: is the win rate significantly above 50%?
+    # p = probability of observing >= n_wins out of n_trades under null (win_rate=0.5).
+    # Low p → strategy has a significant positive edge per trade.
+    from scipy.stats import binomtest
+    n_wins = int((trades > 0).sum())
+    n_total = len(trades)
+    result = binomtest(n_wins, n_total, p=0.5, alternative="greater")
+    return float(result.pvalue)


 def walk_forward_rolling(
@@ -75,8 +75,10 @@ class CoSTEER(Developer[Experiment]):

    def _get_last_fb(self) -> CoSTEERMultiFeedback:
        fb = self.evolve_agent.evolving_trace[-1].feedback
-        assert fb is not None, "feedback is None"
-        assert isinstance(fb, CoSTEERMultiFeedback), "feedback must be of type CoSTEERMultiFeedback"
+        if fb is None:
+            raise AssertionError("feedback is None")
+        if not isinstance(fb, CoSTEERMultiFeedback):
+            raise TypeError("feedback must be of type CoSTEERMultiFeedback")
        return fb

    def should_use_new_evo(self, base_fb: CoSTEERMultiFeedback | None, new_fb: CoSTEERMultiFeedback) -> bool:
@@ -121,7 +123,8 @@ class CoSTEER(Developer[Experiment]):

        for evo_exp in self.evolve_agent.multistep_evolve(evo_exp, self.evaluator):
            iteration_count += 1
-            assert isinstance(evo_exp, Experiment)  # multiple inheritance
+            if not isinstance(evo_exp, Experiment):
+                raise TypeError("evo_exp must be an instance of Experiment")
            evo_fb = self._get_last_fb()
            update_fallback = self.should_use_new_evo(
                base_fb=fallback_evo_fb,
@@ -154,7 +157,8 @@ class CoSTEER(Developer[Experiment]):
                evo_exp = fallback_evo_exp
                evo_exp.recover_ws_ckp()
                evo_fb = fallback_evo_fb
-            assert evo_fb is not None  # multistep_evolve should run at least once
+            if evo_fb is None:
+                raise AssertionError("multistep_evolve should run at least once")
            evo_exp = self._exp_postprocess_by_feedback(evo_exp, evo_fb)
        except CoderError as e:
            e.caused_by_timeout = reached_max_seconds
@@ -264,9 +268,12 @@ class CoSTEER(Developer[Experiment]):
        - Raise Error if it failed to handle the develop task
        -
        """
-        assert isinstance(evo, Experiment)
-        assert isinstance(feedback, CoSTEERMultiFeedback)
-        assert len(evo.sub_workspace_list) == len(feedback)
+        if not isinstance(evo, Experiment):
+            raise TypeError("evo must be an instance of Experiment")
+        if not isinstance(feedback, CoSTEERMultiFeedback):
+            raise TypeError("feedback must be an instance of CoSTEERMultiFeedback")
+        if len(evo.sub_workspace_list) != len(feedback):
+            raise ValueError("Length of sub_workspace_list must match length of feedback")

        # FIXME: when whould the feedback be None?
        failed_feedbacks = [
@@ -122,7 +122,8 @@ class MultiProcessEvolvingStrategy(EvolvingStrategy):
        last_feedback = None
        if len(evolving_trace) > 0:
            last_feedback = evolving_trace[-1].feedback
-            assert isinstance(last_feedback, CoSTEERMultiFeedback)
+            if not isinstance(last_feedback, CoSTEERMultiFeedback):
+                raise TypeError("last_feedback must be of type CoSTEERMultiFeedback")

        # 1.找出需要evolve的task
        to_be_finished_task_index: list[int] = []
@@ -1028,7 +1028,8 @@ class CoSTEERKnowledgeBaseV2(EvolvingKnowledgeBase):

        """
        node_count = len(nodes)
-        assert node_count >= 2, "nodes length must >=2"
+        if node_count < 2:
+            raise ValueError("nodes length must >=2")
        intersection_node_list = []
        if output_intersection_origin:
            origin_list = []
@@ -54,7 +54,8 @@ def get_ds_env(
        ValueError: If the env_type is not recognized.
    """
    conf = DSCoderCoSTEERSettings()
-    assert conf_type in ["kaggle", "mlebench"], f"Unknown conf_type: {conf_type}"
+    if conf_type not in ["kaggle", "mlebench"]:
+        raise ValueError(f"Unknown conf_type: {conf_type}")

    if conf.env_type == "docker":
        env_conf = DSDockerConf() if conf_type == "kaggle" else MLEBDockerConf()
@@ -79,7 +80,8 @@ def get_clear_ws_cmd(stage: Literal["before_training", "before_inference"] = "be
    """
    Clean the files in workspace to a specific stage
    """
-    assert stage in ["before_training", "before_inference"], f"Unknown stage: {stage}"
+    if stage not in ["before_training", "before_inference"]:
+        raise ValueError(f"Unknown stage: {stage}")
    if DS_RD_SETTING.enable_model_dump and stage == "before_training":
        cmd = "rm -r submission.csv scores.csv models trace.log"
    else:
@@ -13,7 +13,7 @@ File structure

 from pathlib import Path

-from jinja2 import Environment, StrictUndefined
+from jinja2 import Environment, StrictUndefined, select_autoescape

 from rdagent.app.data_science.conf import DS_RD_SETTING
 from rdagent.components.coder.CoSTEER.evaluators import (
@@ -88,7 +88,7 @@ class EnsembleMultiProcessEvolvingStrategy(MultiProcessEvolvingStrategy):
            code_spec = workspace.file_dict["spec/ensemble.md"]
        else:
            test_code = (
-                Environment(undefined=StrictUndefined)
+                Environment(undefined=StrictUndefined, autoescape=select_autoescape())
                .from_string((DIRNAME / "eval_tests" / "ensemble_test.txt").read_text())
                .render(
                    model_names=[
@@ -2,7 +2,7 @@ import json
 import re
 from pathlib import Path

-from jinja2 import Environment, StrictUndefined
+from jinja2 import Environment, StrictUndefined, select_autoescape

 from rdagent.app.data_science.conf import DS_RD_SETTING
 from rdagent.components.coder.CoSTEER.evaluators import (
@@ -55,7 +55,7 @@ class EnsembleCoSTEEREvaluator(CoSTEEREvaluator):
        fname = "test/ensemble_test.txt"
        test_code = (DIRNAME / "eval_tests" / "ensemble_test.txt").read_text()
        test_code = (
-            Environment(undefined=StrictUndefined)
+            Environment(undefined=StrictUndefined, autoescape=select_autoescape())
            .from_string(test_code)
            .render(
                model_names=[
@@ -51,13 +51,23 @@ class FactorAutoFixer:
        self.fixes_applied = []
        fixed_code = code

-        # Apply fixes in order - groupby fixes MUST come before min_periods fixes
+        # Apply fixes in order
+        # NOTE: _fix_min_periods is intentionally excluded — it increased min_periods to
+        # match window size, which causes all-NaN output for intraday data with 96 bars/day
+        # (window=240 > 96 means zero valid bars per day). The LLM sets its own min_periods.
        fix_methods = [
-            self._fix_groupby_apply_to_transform,  # First: fix groupby patterns
-            self._fix_min_periods,                  # Second: fix min_periods in resulting rolling calls
-            self._fix_inf_nan_handling,             # Third: add inf/nan handling
-            self._fix_data_range_processing,        # Fourth: ensure full data range
-            self._fix_multiindex_groupby,           # Fifth: ensure groupby on MultiIndex
+            self._fix_instrument_column_access,     # First: fix df['instrument'] on MultiIndex
+            self._fix_instrument_loc_multiindex,    # Second: fix df.loc[instrument_var] on MultiIndex
+            self._fix_zero_volume_proxy,            # Third: replace zero $volume with range proxy
+            self._fix_reset_index_groupby,          # Fourth: fix groupby(level=N) after reset_index()
+            self._fix_groupby_mixed_levels,         # Fifth: fix groupby(level=[int, str])
+            self._fix_groupby_column_on_multiindex, # Sixth: fix groupby(['instrument','date']) on MultiIndex
+            self._fix_chained_groupby,              # Seventh: fix groupby(level=N).groupby('date') chain
+            self._fix_rolling_ddof,                 # Eighth: remove unsupported ddof kwarg
+            self._fix_groupby_apply_to_transform,   # Ninth: fix groupby patterns
+            self._fix_inf_nan_handling,             # Tenth: add inf/nan handling
+            self._fix_data_range_processing,        # Eleventh: ensure full data range
+            self._fix_multiindex_groupby,           # Twelfth: ensure groupby on MultiIndex
        ]

        for fix_method in fix_methods:
@@ -75,6 +85,352 @@ class FactorAutoFixer:

        return fixed_code

+    def _fix_instrument_column_access(self, code: str) -> str:
+        """
+        Fix: df['instrument'] raises KeyError on a MultiIndex DataFrame because
+        'instrument' is an index level (level 1), not a column.
+
+        Replace df['instrument'] with df.index.get_level_values('instrument')
+        but only when the DataFrame has a MultiIndex (not after reset_index which
+        would have promoted it to a real column).
+
+        Also fixes df.reset_index()['instrument'] correctly since after reset_index
+        the column exists.
+        """
+        fixed_code = code
+
+        # Skip if already fixed or if reset_index() is being used before the access
+        # We only fix bare df['instrument'] where df is the original MultiIndex frame.
+        # Heuristic: if the assignment lhs or context shows reset_index, leave it alone.
+
+        # Pattern: <varname>['instrument'] where varname is NOT a reset_index result
+        reset_vars = set(re.findall(r'(\w+)\s*=\s*\w[^=\n]*\.reset_index\(', fixed_code))
+
+        def _replace_instrument_access(m: re.Match) -> str:
+            var = m.group(1)
+            if var in reset_vars:
+                return m.group(0)  # leave reset_index vars alone — column exists
+            self.fixes_applied.append(f"instrument_column: {var}['instrument'] → get_level_values(1)")
+            return f"{var}.index.get_level_values(1)"
+
+        # Exclude assignment targets: var['instrument'] = ... must not become
+        # var.index.get_level_values(1) = ... (SyntaxError: cannot assign to function call)
+        fixed_code = re.sub(r"(\w+)\['instrument'\](?!\s*=)", _replace_instrument_access, fixed_code)
+
+        return fixed_code
+
+    def _fix_instrument_loc_multiindex(self, code: str) -> str:
+        """
+        Fix: df.loc[instrument_var] raises DateParseError on a (datetime, instrument)
+        MultiIndex because pandas tries to match the instrument string against the
+        datetime level (level 0).
+
+        Pattern detected: for-loops iterating over get_level_values('instrument') or
+        get_level_values(1) where the loop variable is then used as df.loc[loop_var].
+
+        Replacement: df.loc[instrument_var] → df.xs(instrument_var, level=1)
+        """
+        fixed_code = code
+
+        # Find variables iterated from get_level_values('instrument') or get_level_values(1)
+        inst_vars = set(
+            re.findall(
+                r"for\s+(\w+)\s+in\s+.+?\.get_level_values\s*\(\s*(?:1|['\"]instrument['\"])\s*\)[^:\n]*:",
+                code,
+            )
+        )
+
+        if not inst_vars:
+            return fixed_code
+
+        for var in inst_vars:
+            # Replace DF.loc[var] (read) with DF.xs(var, level=1)
+            # Exclude write-back patterns (DF.loc[var] = ...) — leave those as-is
+            def _make_replacer(v: str):
+                def _replace(m: re.Match) -> str:
+                    df_var = m.group(1)
+                    self.fixes_applied.append(
+                        f"instrument_loc: {df_var}.loc[{v}] → {df_var}.xs({v}, level=1)"
+                    )
+                    return f"{df_var}.xs({v}, level=1)"
+
+                return _replace
+
+            # Only match when NOT followed by ' =' (assignment)
+            fixed_code = re.sub(
+                rf"(\w+)\.loc\[\s*{re.escape(var)}\s*\](?!\s*=)",
+                _make_replacer(var),
+                fixed_code,
+            )
+
+        return fixed_code
+
+    def _fix_zero_volume_proxy(self, code: str) -> str:
+        """
+        Fix: $volume is always 0 in our EUR/USD dataset (FX has no real volume).
+        Any factor using $volume (VWAP, volume-weighted returns, etc.) produces
+        all-NaN output because 0*price=0 and sum(0)/sum(0)=NaN.
+
+        Insert a guard right after pd.read_hdf() that replaces zero volume with
+        the intraday price-range proxy ($high - $low) so volume-weighted factors
+        produce meaningful signals.
+        """
+        if "'$volume'" not in code and '"$volume"' not in code:
+            return code
+
+        # Already patched
+        if "volume proxy" in code:
+            return code
+
+        lines = code.splitlines()
+        insert_after = -1
+        df_var = "df"
+        indent = "    "
+
+        for i, line in enumerate(lines):
+            if "read_hdf(" in line:
+                m = re.match(r"(\s*)(\w+)\s*=\s*", line)
+                if m:
+                    indent = m.group(1)
+                    df_var = m.group(2)
+                else:
+                    m2 = re.match(r"(\s*)", line)
+                    indent = m2.group(1) if m2 else "    "
+                insert_after = i
+                break
+
+        if insert_after == -1:
+            return code
+
+        proxy_lines = [
+            f"{indent}# volume proxy: $volume is always 0 in FX data — use price-range as proxy",
+            f"{indent}if ({df_var}['$volume'] == 0).all():",
+            f"{indent}    {df_var}['$volume'] = {df_var}['$high'] - {df_var}['$low']",
+        ]
+        lines = lines[: insert_after + 1] + proxy_lines + lines[insert_after + 1 :]
+        self.fixes_applied.append("volume_proxy: replaced zero $volume with ($high - $low)")
+        return "\n".join(lines)
+
+    def _fix_reset_index_groupby(self, code: str) -> str:
+        """
+        Fix: groupby(level=N) on a variable created by .reset_index() fails because
+        reset_index() converts the MultiIndex into regular columns, leaving a plain
+        RangeIndex.  Replace groupby(level=N) on such variables with
+        groupby('instrument').
+
+        Detected pattern:
+            varname = <anything>.reset_index(...)
+            ...
+            varname.groupby(level=0|1)
+        """
+        fixed_code = code
+
+        # Find all variables assigned via reset_index()
+        reset_vars = set(re.findall(r'(\w+)\s*=\s*\w[^=\n]*\.reset_index\(', fixed_code))
+
+        for var in reset_vars:
+            # Replace var.groupby(level=N) with var.groupby('instrument')
+            pattern = rf'{re.escape(var)}\.groupby\(level\s*=\s*\d+\)'
+            if re.search(pattern, fixed_code):
+                fixed_code = re.sub(pattern, f"{var}.groupby('instrument')", fixed_code)
+                self.fixes_applied.append(f"reset_index_groupby: {var}.groupby(level=N) → groupby('instrument')")
+
+        return fixed_code
+
+    def _fix_groupby_mixed_levels(self, code: str) -> str:
+        """
+        Fix: groupby(level=[int, 'str']) raises AssertionError because string level
+        names don't exist on an unnamed MultiIndex.  Keep only integer levels.
+
+        Pattern:  .groupby(level=[0, 'date'])  →  .groupby(level=0)
+                  .groupby(level=[1, 'date'])  →  .groupby(level=1)
+        """
+        fixed_code = code
+
+        def _keep_int_levels(m):
+            inner = m.group(1)
+            ints = re.findall(r'\b(\d+)\b', inner)
+            if not ints:
+                return m.group(0)
+            replacement = f'.groupby(level={ints[0]})' if len(ints) == 1 else f'.groupby(level=[{", ".join(ints)}])'
+            self.fixes_applied.append(f"mixed_levels: groupby(level=[...,str]) → {replacement}")
+            return replacement
+
+        fixed_code = re.sub(r'\.groupby\(level=\[([^\]]+)\]\)', _keep_int_levels, fixed_code)
+        return fixed_code
+
+    def _fix_groupby_column_on_multiindex(self, code: str) -> str:
+        """
+        Fix: groupby(['instrument', 'date']) on a MultiIndex (datetime, instrument)
+        DataFrame fails with KeyError because those are index levels, not columns.
+
+        Correct replacement preserves BOTH dimensions so intraday calculations reset
+        per day:
+            var.groupby(['instrument', 'date'])
+            → var.groupby([var.index.get_level_values(1), var.index.get_level_values(0).normalize()])
+
+        Single-column groupby(['instrument']) is correctly replaced with groupby(level=1).
+        Note: do NOT convert groupby('instrument') → groupby(level=1) here — that would
+        undo the reset_index_groupby fix which correctly emits groupby('instrument').
+        """
+        fixed_code = code
+
+        # Variables created via reset_index() have a plain RangeIndex — applying
+        # get_level_values() on them would raise AttributeError. Skip those.
+        reset_vars = set(re.findall(r'(\w+)\s*=\s*\w[^=\n]*\.reset_index\(', fixed_code))
+
+        def _replace_two_col_groupby(m: re.Match, order: str) -> str:
+            var = m.group(1)
+            if var in reset_vars:
+                return m.group(0)  # leave reset_index vars alone — RangeIndex, not MultiIndex
+            if order == "instrument_date":
+                repl = (
+                    f"{var}.groupby([{var}.index.get_level_values(1), "
+                    f"{var}.index.get_level_values(0).normalize()])"
+                )
+            else:  # date_instrument
+                repl = (
+                    f"{var}.groupby([{var}.index.get_level_values(0).normalize(), "
+                    f"{var}.index.get_level_values(1)])"
+                )
+            self.fixes_applied.append(f"multiindex_groupby: {m.group(0)[:60]} → two-level")
+            return repl
+
+        # groupby(['instrument', 'date']) — capture variable name before .groupby
+        fixed_code = re.sub(
+            r'(\w+)\.groupby\(\[\'instrument\',\s*\'date\'\]\)',
+            lambda m: _replace_two_col_groupby(m, "instrument_date"),
+            fixed_code,
+        )
+        # groupby(['date', 'instrument'])
+        fixed_code = re.sub(
+            r'(\w+)\.groupby\(\[\'date\',\s*\'instrument\'\]\)',
+            lambda m: _replace_two_col_groupby(m, "date_instrument"),
+            fixed_code,
+        )
+        # single: groupby(['instrument']) → groupby(level=1), but not on reset_index vars
+        def _replace_single_instrument_groupby(m: re.Match) -> str:
+            # Look backwards to find the variable name
+            prefix = fixed_code[: m.start()]
+            var_match = re.search(r'(\w+)\s*$', prefix)
+            var = var_match.group(1) if var_match else ''
+            if var in reset_vars:
+                return m.group(0)
+            self.fixes_applied.append("multiindex_groupby: groupby(['instrument']) → groupby(level=1)")
+            return ".groupby(level=1)"
+
+        if re.search(r"\.groupby\(\['instrument'\]\)", fixed_code):
+            fixed_code = re.sub(r"\.groupby\(\['instrument'\]\)", _replace_single_instrument_groupby, fixed_code)
+
+        # groupby(level=['instrument', 'date']) — uses level= keyword with string names.
+        # 'date' is NOT a valid level name in our (datetime, instrument) MultiIndex;
+        # replace with get_level_values to normalize datetime to daily timestamps.
+        fixed_code = re.sub(
+            r"(\w+)\.groupby\(level=\['instrument',\s*'date'\]\)",
+            lambda m: (
+                self.fixes_applied.append(
+                    f"multiindex_groupby: {m.group(0)[:60]} → two-level get_level_values"
+                )
+                or f"{m.group(1)}.groupby([{m.group(1)}.index.get_level_values(1), "
+                f"{m.group(1)}.index.get_level_values(0).normalize()])"
+            ),
+            fixed_code,
+        )
+        # groupby(level=['date', 'instrument'])
+        fixed_code = re.sub(
+            r"(\w+)\.groupby\(level=\['date',\s*'instrument'\]\)",
+            lambda m: (
+                self.fixes_applied.append(
+                    f"multiindex_groupby: {m.group(0)[:60]} → two-level get_level_values"
+                )
+                or f"{m.group(1)}.groupby([{m.group(1)}.index.get_level_values(0).normalize(), "
+                f"{m.group(1)}.index.get_level_values(1)])"
+            ),
+            fixed_code,
+        )
+        # single: groupby(level=['instrument']) → groupby(level=1)
+        fixed_code = re.sub(
+            r"\.groupby\(level=\['instrument'\]\)",
+            lambda m: (self.fixes_applied.append("multiindex_groupby: groupby(level=['instrument']) → level=1") or ".groupby(level=1)"),
+            fixed_code,
+        )
+
+        return fixed_code
+
+    def _fix_chained_groupby(self, code: str) -> str:
+        """
+        Fix two broken patterns the LLM generates when trying to group by (instrument, date):
+
+        Pattern A — chained groupby (runtime AttributeError):
+            var.groupby(level=1).groupby('date')
+            → var.groupby([var.index.get_level_values(1),
+                           var.index.get_level_values(0).normalize()])
+
+        Pattern B — keyword arg inside list (SyntaxError):
+            var.groupby([level=1, 'date'])
+            → same two-level replacement
+        """
+        fixed_code = code
+
+        def _two_level(var: str, tag: str) -> str:
+            self.fixes_applied.append(f"chained_groupby: {tag} → two-level")
+            return (
+                f"{var}.groupby([{var}.index.get_level_values(1), "
+                f"{var}.index.get_level_values(0).normalize()])"
+            )
+
+        # Pattern A: var.groupby(level=N).groupby('date')
+        fixed_code = re.sub(
+            r'(\w+)\.groupby\(level=\d+\)\.groupby\(["\']date["\']\)',
+            lambda m: _two_level(m.group(1), m.group(0)[:60]),
+            fixed_code,
+        )
+
+        # Pattern B: .groupby([level=N, 'date']) — SyntaxError in Python.
+        # The variable before .groupby may be complex (e.g. df[mask]) so we don't
+        # try to capture it; we use df as the index reference (always correct since
+        # all filtered frames share df's MultiIndex structure).
+        def _two_level_df(tag: str) -> str:
+            self.fixes_applied.append(f"chained_groupby: {tag} → two-level")
+            return ".groupby([df.index.get_level_values(1), df.index.get_level_values(0).normalize()])"
+
+        fixed_code = re.sub(
+            r'\.groupby\(\[\s*level\s*=\s*\d+\s*,\s*["\']?date["\']?\s*\]\)',
+            lambda m: _two_level_df(m.group(0)[:60]),
+            fixed_code,
+        )
+        # Also handle reversed order: ['date', level=N]
+        fixed_code = re.sub(
+            r'\.groupby\(\[\s*["\']?date["\']?\s*,\s*level\s*=\s*\d+\s*\]\)',
+            lambda m: _two_level_df(m.group(0)[:60]),
+            fixed_code,
+        )
+
+        return fixed_code
+
+    def _fix_rolling_ddof(self, code: str) -> str:
+        """
+        Fix: pandas rolling() does not accept a ddof kwarg — raises TypeError.
+        Remove ddof from both rolling(..., ddof=N) and rolling(...).std(ddof=N).
+        """
+        fixed_code = code
+
+        # Form 1: ddof inside rolling() — .rolling(window=N, min_periods=M, ddof=K)
+        def _strip_ddof_from_rolling(m):
+            inner = re.sub(r',?\s*ddof\s*=\s*\d+', '', m.group(1))
+            inner = inner.strip(', ')
+            self.fixes_applied.append("rolling_ddof: removed ddof from rolling()")
+            return f'.rolling({inner})'
+
+        fixed_code = re.sub(r'\.rolling\(([^)]*ddof\s*=\s*\d+[^)]*)\)', _strip_ddof_from_rolling, fixed_code)
+
+        # Form 2: ddof inside .std() / .var() — .std(ddof=N)
+        if re.search(r'\.(std|var)\([^)]*ddof\s*=\s*\d+', fixed_code):
+            fixed_code = re.sub(r'\.(std|var)\([^)]*ddof\s*=\s*\d+[^)]*\)', r'.\1()', fixed_code)
+            self.fixes_applied.append("rolling_ddof: removed ddof from std()/var()")
+
+        return fixed_code
+
    def _fix_min_periods(self, code: str) -> str:
        """
        Fix: Ensure min_periods matches window size in rolling calculations.
@@ -325,6 +681,45 @@ class FactorAutoFixer:
            fixed_code = fixed_code.replace(old_code, new_code)
            self.fixes_applied.append(f"groupby: fixed rolling correlation (window={window}) with reset_index")

+        # === GENERAL FIX: DF.groupby(level=N)['col'].apply(lambda x: EXPR) ===
+        # apply() on a grouped Series returns a MultiIndex result (extra level prepended),
+        # causing index shape mismatch when assigned back to df['col'].
+        # Replace with transform() which preserves the original index.
+        col_apply_pattern = re.compile(
+            r"(\w+)\.groupby\(level=(\d+)\)\['([^']+)'\]\.apply\((\s*lambda\s+\w+\s*:.*?)\)",
+            re.DOTALL,
+        )
+        for m in list(col_apply_pattern.finditer(fixed_code)):
+            full = m.group(0)
+            df_var = m.group(1)
+            level = m.group(2)
+            col = m.group(3)
+            lam = m.group(4).strip()
+            new_expr = f"{df_var}.groupby(level={level})['{col}'].transform({lam})"
+            fixed_code = fixed_code.replace(full, new_expr, 1)
+            self.fixes_applied.append(
+                f"groupby: {df_var}.groupby(level={level})['{col}'].apply() → transform()"
+            )
+
+        # === FIX: .transform(...).reset_index(level=N, drop=True) ===
+        # transform() already returns the same index as the input — adding reset_index()
+        # after it drops an index level and causes ValueError on assignment back to df['col'].
+        # Detected line-by-line: if a line contains both .transform( and .reset_index(level=
+        reset_suffix = re.compile(r'\s*\.reset_index\s*\(\s*level\s*=[^,)]+,\s*drop\s*=\s*True\s*\)\s*$')
+        new_lines = []
+        changed = False
+        for line in fixed_code.splitlines():
+            if '.transform(' in line and '.reset_index(' in line:
+                cleaned = reset_suffix.sub('', line)
+                if cleaned != line:
+                    new_lines.append(cleaned)
+                    changed = True
+                    continue
+            new_lines.append(line)
+        if changed:
+            fixed_code = '\n'.join(new_lines)
+            self.fixes_applied.append("groupby: removed spurious .reset_index() after .transform()")
+
        # Pattern: Simple groupby().apply() with rolling().method()
        # df.groupby(level=N).apply(lambda x: x['col'].rolling(...).method())
        apply_pattern = r"df\.groupby\(level=(\d+)\)\.apply\(\s*lambda\s+x:\s+x\['([^']+)'\]\.rolling\([^)]+\)\.(\w+)\([^)]*\)\s*\)"
@@ -161,8 +161,7 @@ class FactorFBWorkspace(FBWorkspace):

            try:
                subprocess.check_output(
-                    f"{FACTOR_COSTEER_SETTINGS.python_bin} {execution_code_path}",
-                    shell=True,
+                    [FACTOR_COSTEER_SETTINGS.python_bin, str(execution_code_path)],
                    cwd=self.workspace_path,
                    stderr=subprocess.STDOUT,
                    timeout=FACTOR_COSTEER_SETTINGS.file_based_execution_timeout,
@@ -53,7 +53,7 @@ evolving_strategy_factor_implementation_v1_system: |-
  - ALWAYS use `min_periods=N` where N equals the window size in rolling calculations (e.g., `.rolling(20, min_periods=20)`)
  - ALWAYS handle infinite values after division: `.replace([np.inf, -np.inf], np.nan)` before saving results
  - ALWAYS use `groupby(level=1)` or `groupby('instrument')` before rolling operations on MultiIndex dataframes
-  - Process the COMPLETE date range (2020-2026), do NOT filter by date
+  - Process the COMPLETE date range available in the HDF5 file (do NOT filter by date — the file may contain 2024 debug data or full 2020-2026 data)
  - Use `groupby().transform()` instead of `groupby().apply()` for single-column assignments

  Notice that you should not add any other text before or after the json format.
@@ -6,6 +6,7 @@ Two-step validation:
 2. Micro-batch testing - Runtime validation with small dataset
 """

+import ast
 import json
 import re
 import time
@@ -229,7 +230,7 @@ class LLMConfigValidator:
            final_metrics = re.search(r"\{'train_runtime':[^}]+\}", stdout)
            if final_metrics:
                try:
-                    metrics = eval(final_metrics.group(0))  # Safe: only numbers and strings
+                    metrics = ast.literal_eval(final_metrics.group(0))
                    result["final_metrics"] = {
                        "train_loss": metrics.get("train_loss"),
                        "train_runtime": metrics.get("train_runtime"),
@@ -58,10 +58,12 @@ class ModelCodeEvaluator(CoSTEEREvaluator):
        model_execution_feedback: str = "",
        model_value_feedback: str = "",
    ):
-        assert isinstance(target_task, ModelTask)
-        assert isinstance(implementation, ModelFBWorkspace)
-        if gt_implementation is not None:
-            assert isinstance(gt_implementation, ModelFBWorkspace)
+        if not isinstance(target_task, ModelTask):
+            raise TypeError("target_task must be of type ModelTask")
+        if not isinstance(implementation, ModelFBWorkspace):
+            raise TypeError("implementation must be of type ModelFBWorkspace")
+        if gt_implementation is not None and not isinstance(gt_implementation, ModelFBWorkspace):
+            raise TypeError("gt_implementation must be of type ModelFBWorkspace")

        model_task_information = target_task.get_task_information()
        code = implementation.all_codes
@@ -113,10 +115,12 @@ class ModelFinalEvaluator(CoSTEEREvaluator):
        model_value_feedback: str,
        model_code_feedback: str,
    ):
-        assert isinstance(target_task, ModelTask)
-        assert isinstance(implementation, ModelFBWorkspace)
-        if gt_implementation is not None:
-            assert isinstance(gt_implementation, ModelFBWorkspace)
+        if not isinstance(target_task, ModelTask):
+            raise TypeError("target_task must be of type ModelTask")
+        if not isinstance(implementation, ModelFBWorkspace):
+            raise TypeError("implementation must be of type ModelFBWorkspace")
+        if gt_implementation is not None and not isinstance(gt_implementation, ModelFBWorkspace):
+            raise TypeError("gt_implementation must be of type ModelFBWorkspace")

        system_prompt = T(".prompts:evaluator_final_feedback.system").r(
            scenario=(
@@ -41,7 +41,8 @@ class ModelCoSTEEREvaluator(CoSTEEREvaluator):
                final_feedback="This task has failed too many times, skip implementation.",
                final_decision=False,
            )
-        assert isinstance(target_task, ModelTask)
+        if not isinstance(target_task, ModelTask):
+            raise TypeError(f"Expected ModelTask, got {type(target_task)}")

        # NOTE: Use fixed input to test the model to avoid randomness
        batch_size = 8
@@ -50,7 +51,8 @@ class ModelCoSTEEREvaluator(CoSTEEREvaluator):
        input_value = 0.4
        param_init_value = 0.6

-        assert isinstance(implementation, ModelFBWorkspace)
+        if not isinstance(implementation, ModelFBWorkspace):
+            raise TypeError(f"Expected ModelFBWorkspace, got {type(implementation)}")
        model_execution_feedback, gen_np_array = implementation.execute(
            batch_size=batch_size,
            num_features=num_features,
@@ -59,7 +61,8 @@ class ModelCoSTEEREvaluator(CoSTEEREvaluator):
            param_init_value=param_init_value,
        )
        if gt_implementation is not None:
-            assert isinstance(gt_implementation, ModelFBWorkspace)
+            if not isinstance(gt_implementation, ModelFBWorkspace):
+                raise TypeError(f"Expected ModelFBWorkspace, got {type(gt_implementation)}")
            _, gt_np_array = gt_implementation.execute(
                batch_size=batch_size,
                num_features=num_features,
@@ -85,13 +85,16 @@ def load_and_process_one_pdf_by_azure_document_intelligence(


 def load_and_process_pdfs_by_azure_document_intelligence(path: Path) -> dict[str, str]:
-    assert RD_AGENT_SETTINGS.azure_document_intelligence_key is not None
-    assert RD_AGENT_SETTINGS.azure_document_intelligence_endpoint is not None
+    if RD_AGENT_SETTINGS.azure_document_intelligence_key is None:
+        raise AssertionError("azure_document_intelligence_key must be set")
+    if RD_AGENT_SETTINGS.azure_document_intelligence_endpoint is None:
+        raise AssertionError("azure_document_intelligence_endpoint must be set")

    content_dict = {}
    ab_path = path.resolve()
    if ab_path.is_file():
-        assert ".pdf" in ab_path.suffixes, "The file must be a PDF file."
+        if ".pdf" not in ab_path.suffixes:
+            raise ValueError("The file must be a PDF file.")
        proc = load_and_process_one_pdf_by_azure_document_intelligence
        content_dict[str(ab_path)] = proc(
            ab_path,
@@ -24,7 +24,8 @@ class UndirectedNode(Node):
        super().__init__(content, label, embedding)
        self.neighbors: set[UndirectedNode] = set()
        self.appendix = appendix  # appendix stores any additional information
-        assert isinstance(content, str), "content must be a string"
+        if not isinstance(content, str):
+            raise TypeError("content must be a string")

    def add_neighbor(self, node: UndirectedNode) -> None:
        self.neighbors.add(node)
@@ -96,7 +97,8 @@ class Graph(KnowledgeBase):
                APIBackend().create_embedding(input_content=contents[i : i + size]),
            )

-        assert len(nodes) == len(embeddings), "nodes' length must equals embeddings' length"
+        if len(nodes) != len(embeddings):
+            raise ValueError("nodes' length must equal embeddings' length")
        for node, embedding in zip(nodes, embeddings):
            node.embedding = embedding
        return nodes
@@ -252,7 +254,8 @@ class UndirectedGraph(Graph):

        """
        min_nodes_count = 2
-        assert len(nodes) >= min_nodes_count, "nodes length must >=2"
+        if len(nodes) < min_nodes_count:
+            raise ValueError("nodes length must >=2")
        intersection = None

        for node in nodes:
@@ -87,7 +87,8 @@ class ModelWsLoader(WsLoader[ModelTask, ModelFBWorkspace]):
        self.path = Path(path)

    def load(self, task: ModelTask) -> ModelFBWorkspace:
-        assert task.name is not None
+        if task.name is None:
+            raise AssertionError("task.name should not be None")
        mti = ModelFBWorkspace(task)
        mti.prepare()
        with open(self.path / f"{task.name}.py", "r") as f:
@@ -4,6 +4,7 @@ import functools
 import importlib
 import json
 import multiprocessing as mp
+import os
 import pickle
 import random
 from collections.abc import Callable
@@ -208,3 +209,13 @@ def cache_with_pickle(hash_func: Callable, post_process_func: Callable | None =
        return cache_wrapper

    return cache_decorator
+
+
+def safe_resolve_path(user_path: Path, safe_root: Path | None = None) -> Path:
+    if safe_root is not None:
+        root_real = os.path.realpath(str(safe_root.expanduser()))
+        path_real = os.path.realpath(str(user_path.expanduser()))  # nosec B614 — validated against safe_root below
+        if not (path_real == root_real or path_real.startswith(root_real + os.sep)):
+            raise ValueError(f"Path {user_path} resolves to {path_real}, outside allowed root {safe_root}")
+        return Path(path_real)
+    return user_path.expanduser().resolve()
@@ -27,6 +27,7 @@ Usage:
 from __future__ import annotations

 import json as _json
+import logging
 import sys
 import threading
 from contextlib import contextmanager
@@ -36,21 +37,24 @@ from typing import Any

 from loguru import logger as _root

-# ── paths ─────────────────────────────────────────────────────────────────────
+# ── paths ─────────────────────────────────────────────────────────────────────────────────
 LOGS_ROOT: Path = Path(__file__).parent.parent.parent / "logs"

-# ── format ────────────────────────────────────────────────────────────────────
+# ── format ────────────────────────────────────────────────────────────────────────────────
 _FILE_FMT = (
    "{time:YYYY-MM-DD HH:mm:ss.SSS} | {level: <8} | {extra[cmd]: <18} | {message}"
 )

-# ── internal state ─────────────────────────────────────────────────────────────
+# ── internal state ─────────────────────────────────────────────────────────────────────────────
 _registered: set[str] = set()   # command keys that already have a file sink
 _all_added: bool = False         # whether the combined all.log sink is active
 _llm_log_lock = threading.Lock()  # guards concurrent writes to llm_calls.jsonl

+# Maximum characters stored per field in llm_calls.jsonl to prevent GB-scale files.
+_LLM_CALL_MAX_CHARS = 500

-# ── helpers ───────────────────────────────────────────────────────────────────
+
+# ── helpers ────────────────────────────────────────────────────────────────────────────────

 def _today_dir() -> Path:
    d = LOGS_ROOT / datetime.now().strftime("%Y-%m-%d")
@@ -79,7 +83,7 @@ def _banner(log, title: str, meta: dict[str, Any]) -> None:
    log.info(sep)


-# ── public API ────────────────────────────────────────────────────────────────
+# ── public API ──────────────────────────────────────────────────────────────────────────────

 def log_llm_call(
    system: str | None,
@@ -88,16 +92,19 @@ def log_llm_call(
    start_time: Any = None,
    end_time: Any = None,
 ) -> None:
-    """Append one complete LLM call to logs/YYYY-MM-DD/llm_calls.jsonl.
+    """Append one LLM call summary to logs/YYYY-MM-DD/llm_calls.jsonl.
+
+    Prompt/response content is capped at _LLM_CALL_MAX_CHARS to prevent
+    GB-scale log files from long-running loops.

    Each line is a self-contained JSON object so the file is grep/jq-friendly:
      jq 'select(.duration_ms > 5000)' logs/2026-04-17/llm_calls.jsonl
    """
    entry: dict[str, Any] = {
        "ts": datetime.now().isoformat(timespec="milliseconds"),
-        "system": system or "",
-        "user": user,
-        "response": response,
+        "system": (system or "")[:_LLM_CALL_MAX_CHARS],
+        "user": user[:_LLM_CALL_MAX_CHARS],
+        "response": response[:_LLM_CALL_MAX_CHARS],
    }
    if start_time is not None and end_time is not None:
        try:
@@ -130,13 +137,13 @@ def setup(command: str, **context: Any):
    key = command.lower()

    if key not in _registered:
-        # Per-command rotating file
        _root.add(
            str(log_dir / f"{key}.log"),
            format=_FILE_FMT,
            filter=lambda r, k=key: r["extra"].get("cmd", "").lower() == k,
-            rotation="00:00",      # new file at midnight
-            retention="30 days",
+            rotation="50 MB",
+            compression="gz",
+            retention="7 days",
            encoding="utf-8",
            enqueue=True,
            backtrace=False,
@@ -145,13 +152,13 @@ def setup(command: str, **context: Any):
        _registered.add(key)

    if not _all_added:
-        # Combined log — all commands
        _root.add(
            str(log_dir / "all.log"),
            format=_FILE_FMT,
            filter=lambda r: "cmd" in r["extra"],
-            rotation="00:00",
-            retention="60 days",
+            rotation="100 MB",
+            compression="gz",
+            retention="7 days",
            encoding="utf-8",
            enqueue=True,
            backtrace=False,
@@ -0,0 +1,306 @@
+import argparse
+import json
+import pickle  # nosec
+import re
+import time
+from pathlib import Path
+
+import streamlit as st
+from streamlit import session_state
+
+from rdagent.log.ui.conf import UI_SETTING
+from rdagent.log.utils import extract_evoid, extract_loopid_func_name
+
+st.set_page_config(layout="wide", page_title="debug_llm", page_icon="🎓", initial_sidebar_state="expanded")
+
+# 获取 log_path 参数
+parser = argparse.ArgumentParser(description="RD-Agent Streamlit App")
+parser.add_argument("--log_dir", type=str, help="Path to the log directory")
+args = parser.parse_args()
+
+
+def get_folders_sorted(log_path):
+    """缓存并返回排序后的文件夹列表，并加入进度打印"""
+    with st.spinner("正在加载文件夹列表..."):
+        folders = sorted(
+            (folder for folder in log_path.iterdir() if folder.is_dir() and list(folder.iterdir())),
+            key=lambda folder: folder.stat().st_mtime,
+            reverse=True,
+        )
+        st.write(f"找到 {len(folders)} 个文件夹")
+    return [folder.name for folder in folders]
+
+
+if UI_SETTING.enable_cache:
+    get_folders_sorted = st.cache_data(get_folders_sorted)
+
+
+# 设置主日志路径
+main_log_path = Path(args.log_dir) if args.log_dir else Path("./log")
+if not main_log_path.exists():
+    st.error(f"Log dir {main_log_path} does not exist!")
+    st.stop()
+
+if "data" not in session_state:
+    session_state.data = []
+if "log_path" not in session_state:
+    session_state.log_path = None
+
+tlist = []
+
+
+def load_data():
+    """加载数据到 session_state 并显示进度"""
+    log_file = main_log_path / session_state.log_path / "debug_llm.pkl"
+    try:
+        with st.spinner(f"正在加载数据文件 {log_file}..."):
+            start_time = time.time()
+            with open(log_file, "rb") as f:
+                session_state.data = pickle.load(f, encoding="utf-8")  # nosec
+            st.success(f"数据加载完成！耗时 {time.time() - start_time:.2f} 秒")
+            st.session_state["current_loop"] = 1
+    except Exception as e:
+        session_state.data = [{"error": str(e)}]
+        st.error(f"加载数据失败: {e}")
+
+
+# UI - Sidebar
+with st.sidebar:
+    st.markdown(":blue[**Log Path**]")
+    manually = st.toggle("Manual Input")
+    if manually:
+        st.text_input("log path", key="log_path", label_visibility="collapsed")
+    else:
+        folders = get_folders_sorted(main_log_path)
+        st.selectbox(f"**Select from {main_log_path.absolute()}**", folders, key="log_path")  # nosec B608 — not SQL, Bandit false positive on "Select" in UI label
+
+    if st.button("Refresh Data"):
+        load_data()
+        st.rerun()
+
+
+# Helper functions
+def show_text(text, lang=None):
+    """显示文本代码块"""
+    if lang:
+        st.code(text, language=lang, wrap_lines=True)
+    elif "\n" in text:
+        st.code(text, language="python", wrap_lines=True)
+    else:
+        st.code(text, language="html", wrap_lines=True)
+
+
+def highlight_prompts_uri(uri):
+    """高亮 URI 的格式"""
+    parts = uri.split(":")
+    return f"**{parts[0]}:**:green[**{parts[1]}**]"
+
+
+# Display Data
+progress_text = st.empty()
+progress_bar = st.progress(0)
+
+# 每页展示一个 Loop
+LOOPS_PER_PAGE = 1
+
+# 获取所有的 Loop ID
+loop_groups = {}
+for i, d in enumerate(session_state.data):
+    tag = d["tag"]
+    loop_id, _ = extract_loopid_func_name(tag)
+    if loop_id:
+        if loop_id not in loop_groups:
+            loop_groups[loop_id] = []
+        loop_groups[loop_id].append(d)
+
+# 按 Loop ID 排序
+sorted_loop_ids = sorted(loop_groups.keys(), key=int)  # 假设 Loop ID 是数字
+total_loops = len(sorted_loop_ids)
+total_pages = total_loops  # 每页展示一个 Loop
+
+
+# simple display
+# FIXME: Delete this simple UI if trace have tag(evo_id & loop_id)
+# with st.sidebar:
+#     start = int(st.text_input("start", 0))
+#     end = int(st.text_input("end", 100))
+# for m in session_state.data[start:end]:
+#     if "tpl" in m["tag"]:
+#         obj = m["obj"]
+#         uri = obj["uri"]
+#         tpl = obj["template"]
+#         cxt = obj["context"]
+#         rd = obj["rendered"]
+#         with st.expander(highlight_prompts_uri(uri), expanded=False, icon="⚙️"):
+#             t1, t2, t3 = st.tabs([":green[**Rendered**]", ":blue[**Template**]", ":orange[**Context**]"])
+#             with t1:
+#                 show_text(rd)
+#             with t2:
+#                 show_text(tpl, lang="django")
+#             with t3:
+#                 st.json(cxt)
+#     if "llm" in m["tag"]:
+#         obj = m["obj"]
+#         system = obj.get("system", None)
+#         user = obj["user"]
+#         resp = obj["resp"]
+#         with st.expander(f"**LLM**", expanded=False, icon="🤖"):
+#             t1, t2, t3 = st.tabs([":green[**Response**]", ":blue[**User**]", ":orange[**System**]"])
+#             with t1:
+#                 try:
+#                     rdict = json.loads(resp)
+#                     if "code" in rdict:
+#                         code = rdict["code"]
+#                         st.markdown(":red[**Code in response dict:**]")
+#                         st.code(code, language="python", wrap_lines=True, line_numbers=True)
+#                         rdict.pop("code")
+#                     elif "spec" in rdict:
+#                         spec = rdict["spec"]
+#                         st.markdown(":red[**Spec in response dict:**]")
+#                         st.markdown(spec)
+#                         rdict.pop("spec")
+#                     else:
+#                         # show model codes
+#                         showed_keys = []
+#                         for k, v in rdict.items():
+#                             if k.startswith("model_") and k.endswith(".py"):
+#                                 st.markdown(f":red[**{k}**]")
+#                                 st.code(v, language="python", wrap_lines=True, line_numbers=True)
+#                                 showed_keys.append(k)
+#                         for k in showed_keys:
+#                             rdict.pop(k)
+#                     st.write(":red[**Other parts (except for the code or spec) in response dict:**]")
+#                     st.json(rdict)
+#                 except:
+#                     st.json(resp)
+#             with t2:
+#                 show_text(user)
+#             with t3:
+#                 show_text(system or "No system prompt available")
+
+
+if total_pages:
+    # 初始化 current_loop
+    if "current_loop" not in st.session_state:
+        st.session_state["current_loop"] = 1
+
+    # Loop 导航按钮
+    col1, col2, col3, col4, col5 = st.sidebar.columns([1.2, 1, 2, 1, 1.2])
+
+    with col1:
+        if st.button("|<"):  # 首页
+            st.session_state["current_loop"] = 1
+    with col2:
+        if st.button("<") and st.session_state["current_loop"] > 1:  # 上一页
+            st.session_state["current_loop"] -= 1
+    with col3:
+        # 下拉列表显示所有 Loop
+        st.session_state["current_loop"] = st.selectbox(
+            "选择 Loop",
+            options=list(range(1, total_loops + 1)),
+            index=st.session_state["current_loop"] - 1,  # 默认选中当前 Loop
+            label_visibility="collapsed",  # 隐藏标签
+        )
+    with col4:
+        if st.button("\>") and st.session_state["current_loop"] < total_loops:  # 下一页
+            st.session_state["current_loop"] += 1
+    with col5:
+        if st.button("\>|"):  # 最后一页
+            st.session_state["current_loop"] = total_loops
+
+    # 获取当前 Loop
+    current_loop = st.session_state["current_loop"]
+
+    # 渲染当前 Loop 数据
+    loop_id = sorted_loop_ids[current_loop - 1]
+    progress_text = st.empty()
+    progress_text.text(f"正在处理 Loop {loop_id}...")
+    progress_bar.progress(current_loop / total_loops, text=f"Loop :green[**{current_loop}**] / {total_loops}")
+
+    # 渲染 Loop Header
+    loop_anchor = f"Loop_{loop_id}"
+    if loop_anchor not in tlist:
+        tlist.append(loop_anchor)
+        st.header(loop_anchor, anchor=loop_anchor, divider="blue")
+
+    # 渲染当前 Loop 的所有数据
+    loop_data = loop_groups[loop_id]
+    for d in loop_data:
+        tag = d["tag"]
+        obj = d["obj"]
+        _, func_name = extract_loopid_func_name(tag)
+        evo_id = extract_evoid(tag)
+
+        func_anchor = f"loop_{loop_id}.{func_name}"
+        if func_anchor not in tlist:
+            tlist.append(func_anchor)
+            st.header(f"in *{func_name}*", anchor=func_anchor, divider="green")
+
+        evo_anchor = f"loop_{loop_id}.evo_step_{evo_id}"
+        if evo_id and evo_anchor not in tlist:
+            tlist.append(evo_anchor)
+            st.subheader(f"evo_step_{evo_id}", anchor=evo_anchor, divider="orange")
+
+        # 根据 tag 渲染内容
+        if "debug_exp_gen" in tag:
+            with st.expander(
+                f"Exp in :violet[**{obj.experiment_workspace.workspace_path}**]", expanded=False, icon="🧩"
+            ):
+                st.write(obj)
+        elif "debug_tpl" in tag:
+            uri = obj["uri"]
+            tpl = obj["template"]
+            cxt = obj["context"]
+            rd = obj["rendered"]
+            with st.expander(highlight_prompts_uri(uri), expanded=False, icon="⚙️"):
+                t1, t2, t3 = st.tabs([":green[**Rendered**]", ":blue[**Template**]", ":orange[**Context**]"])
+                with t1:
+                    show_text(rd)
+                with t2:
+                    show_text(tpl, lang="django")
+                with t3:
+                    st.json(cxt)
+        elif "debug_llm" in tag:
+            system = obj.get("system", None)
+            user = obj["user"]
+            resp = obj["resp"]
+            with st.expander(f"**LLM**", expanded=False, icon="🤖"):
+                t1, t2, t3 = st.tabs([":green[**Response**]", ":blue[**User**]", ":orange[**System**]"])
+                with t1:
+                    try:
+                        rdict = json.loads(resp)
+                        if "code" in rdict:
+                            code = rdict["code"]
+                            st.markdown(":red[**Code in response dict:**]")
+                            st.code(code, language="python", wrap_lines=True, line_numbers=True)
+                            rdict.pop("code")
+                        elif "spec" in rdict:
+                            spec = rdict["spec"]
+                            st.markdown(":red[**Spec in response dict:**]")
+                            st.markdown(spec)
+                            rdict.pop("spec")
+                        else:
+                            # show model codes
+                            showed_keys = []
+                            for k, v in rdict.items():
+                                if k.startswith("model_") and k.endswith(".py"):
+                                    st.markdown(f":red[**{k}**]")
+                                    st.code(v, language="python", wrap_lines=True, line_numbers=True)
+                                    showed_keys.append(k)
+                            for k in showed_keys:
+                                rdict.pop(k)
+                        st.write(":red[**Other parts (except for the code or spec) in response dict:**]")
+                        st.json(rdict)
+                    except:
+                        st.json(resp)
+                with t2:
+                    show_text(user)
+                with t3:
+                    show_text(system or "No system prompt available")
+
+    progress_text.text("当前 Loop 数据处理完成！")
+
+    # Sidebar TOC
+    with st.sidebar:
+        toc = "\n".join([f"- [{t}](#{t})" if t.startswith("L") else f"  - [{t.split('.')[1]}](#{t})" for t in tlist])
+        st.markdown(toc, unsafe_allow_html=True)
@@ -541,7 +541,8 @@ class APIBackend(ABC):
        **kwargs,
    ) -> str | list[list[float]]:
        """This function to share operation between embedding and chat completion"""
-        assert not (chat_completion and embedding), "chat_completion and embedding cannot be True at the same time"
+        if chat_completion and embedding:
+            raise ValueError("chat_completion and embedding cannot be True at the same time")
        max_retry = LLM_SETTINGS.max_retry if LLM_SETTINGS.max_retry is not None else max_retry
        timeout_count = 0
        violation_count = 0
@@ -720,7 +721,13 @@ class APIBackend(ABC):

            if finish_reason is None or finish_reason != "length":
                break  # we get a full response now.
-            new_messages.append({"role": "assistant", "content": response})
+            # Merge into the previous assistant message if there already is one at the end.
+            # Appending a second consecutive assistant message causes llama-server to return 400
+            # ("Cannot have 2 or more assistant messages at the end of the list").
+            if new_messages and new_messages[-1]["role"] == "assistant":
+                new_messages[-1]["content"] += response
+            else:
+                new_messages.append({"role": "assistant", "content": response})
        else:
            raise RuntimeError(f"Failed to continue the conversation after {try_n} retries.")

@@ -36,16 +36,18 @@ def get_agent_model() -> OpenAIChatModel:

    """
    backend = APIBackend()
-    assert isinstance(backend, LiteLLMAPIBackend), "Only LiteLLMAPIBackend is supported"
+    if not isinstance(backend, LiteLLMAPIBackend):
+        raise TypeError("Only LiteLLMAPIBackend is supported")

    compl_kwargs = backend.get_complete_kwargs()

    selected_model = compl_kwargs["model"]

    _, custom_llm_provider, _, _ = get_llm_provider(selected_model)
-    assert (
-        custom_llm_provider in PROVIDER_TO_ENV_MAP
-    ), f"Provider {custom_llm_provider} not supported. Please add it into `PROVIDER_TO_ENV_MAP`"
+    if custom_llm_provider not in PROVIDER_TO_ENV_MAP:
+        raise ValueError(
+            f"Provider {custom_llm_provider} not supported. Please add it into `PROVIDER_TO_ENV_MAP`"
+        )
    prefix = PROVIDER_TO_ENV_MAP[custom_llm_provider]
    api_key = os.getenv(f"{prefix}_API_KEY", None)
    api_base = os.getenv(f"{prefix}_API_BASE", None)
@@ -268,7 +268,8 @@ class JsonReducer(DataReducer):
                parent[key] = sampled  # type: ignore  # parent 是 list，key 是 index, list.__setitem__(key, sampled)
            self.sampled_files.extend([self.extract_filename(i) for i in sampled])
            break
-        assert len(self.sampled_files) > 0
+        if len(self.sampled_files) <= 0:
+            raise AssertionError("sampled_files must contain at least one file")
        return data

    def _find_all_lists(
@@ -7,8 +7,10 @@ from sklearn.metrics import roc_auc_score
 def prepare_for_auroc_metric(submission: pd.DataFrame, answers: pd.DataFrame, id_col: str, target_col: str) -> dict:

    # Answers checks
-    assert id_col in answers.columns, f"answers dataframe should have an {id_col} column"
-    assert target_col in answers.columns, f"answers dataframe should have a {target_col} column"
+    if id_col not in answers.columns:
+        raise InvalidSubmissionError(f"answers dataframe should have an {id_col} column")
+    if target_col not in answers.columns:
+        raise InvalidSubmissionError(f"answers dataframe should have a {target_col} column")

    # Submission checks
    if id_col not in submission.columns:
@@ -1,7 +1,8 @@
 from pathlib import Path

 # Check if our submission file exists
-assert Path("submission.csv").exists(), "Error: submission.csv not found"
+if not Path("submission.csv").exists():
+    raise FileNotFoundError("Error: submission.csv not found")

 submission_lines = Path("submission.csv").read_text().splitlines()
 test_lines = Path("submission_test.csv").read_text().splitlines()
@@ -22,7 +22,8 @@ def prepare_for_metric(submission: pd.DataFrame, answers: pd.DataFrame) -> dict:
    if "price" not in submission.columns:
        raise InvalidSubmissionError("Submission DataFrame must contain 'price' columns.")

-    assert "price" in answers.columns, "Answers DataFrame must contain 'price' columns."
+    if "price" not in answers.columns:
+        raise InvalidSubmissionError("Answers DataFrame must contain 'price' columns.")

    if len(submission) != len(answers):
        raise InvalidSubmissionError("Submission must be the same length as the answers.")
@@ -1,7 +1,8 @@
 from pathlib import Path

 # Check if our submission file exists
-assert Path("submission.csv").exists(), "Error: submission.csv not found"
+if not Path("submission.csv").exists():
+    raise FileNotFoundError("Error: submission.csv not found")

 submission_lines = Path("submission.csv").read_text().splitlines()  # 自动生成的
 test_lines = Path("submission_test.csv").read_text().splitlines()  # test.csv
@@ -56,14 +56,17 @@ sparse.save_npz(public / "test" / "X.npz", X_test)
 sparse.save_npz(public / "train" / "X.npz", X_train)
 df_train.to_csv(public / "train" / "ARF_12h.csv", index=False)

-assert (
-    X_train.shape[0] == df_train.shape[0]
-), f"Mismatch: X_train rows ({X_train.shape[0]}) != df_train rows ({df_train.shape[0]})"
-assert (
-    X_test.shape[0] == df_test.shape[0]
-), f"Mismatch: X_test rows ({X_test.shape[0]}) != df_test rows ({df_test.shape[0]})"
-assert df_test.shape[1] == 2, "Public test set should have 2 columns"
-assert df_train.shape[1] == 3, "Public train set should have 3 columns"
-assert len(df_train) + len(df_test) == len(
-    df_label
-), "Length of new_train and new_test should equal length of old_train"
+if X_train.shape[0] != df_train.shape[0]:
+    raise ValueError(
+        f"Mismatch: X_train rows ({X_train.shape[0]}) != df_train rows ({df_train.shape[0]})"
+    )
+if X_test.shape[0] != df_test.shape[0]:
+    raise ValueError(
+        f"Mismatch: X_test rows ({X_test.shape[0]}) != df_test rows ({df_test.shape[0]})"
+    )
+if df_test.shape[1] != 2:
+    raise ValueError("Public test set should have 2 columns")
+if df_train.shape[1] != 3:
+    raise ValueError("Public train set should have 3 columns")
+if len(df_train) + len(df_test) != len(df_label):
+    raise ValueError("Length of new_train and new_test should equal length of old_train")
@@ -25,11 +25,12 @@ def prepare(raw: Path, public: Path, private: Path):
    new_test.to_csv(public / "test.csv", index=False)

    # Checks
-    assert new_test.shape[1] == 12, "Public test set should have 12 columns"
-    assert new_train.shape[1] == 13, "Public train set should have 13 columns"
-    assert len(new_train) + len(new_test) == len(
-        old_train
-    ), "Length of new_train and new_test should equal length of old_train"
+    if new_test.shape[1] != 12:
+        raise AssertionError("Public test set should have 12 columns")
+    if new_train.shape[1] != 13:
+        raise AssertionError("Public train set should have 13 columns")
+    if len(new_train) + len(new_test) != len(old_train):
+        raise AssertionError("Length of new_train and new_test should equal length of old_train")


 if __name__ == "__main__":
@@ -320,7 +320,8 @@ class DataScienceRDLoop(RDLoop):
            # only clean current workspace without affecting other loops.
            for k in "direct_exp_gen", "coding", "running":
                if k in prev_out and prev_out[k] is not None:
-                    assert isinstance(prev_out[k], DSExperiment)
+                    if not isinstance(prev_out[k], DSExperiment):
+                        raise TypeError(f"prev_out[{k!r}] must be an instance of DSExperiment")
                    clean_workspace(prev_out[k].experiment_workspace.workspace_path)

            # Backup the workspace (only necessary files are included)
@@ -213,7 +213,8 @@ class DSTrace(Trace[DataScienceScen, KnowledgeBase]):
        self, component: COMPONENT, search_list: list[tuple[DSExperiment, ExperimentFeedback]] = []
    ) -> bool:
        for exp, fb in search_list:
-            assert isinstance(exp.hypothesis, DSHypothesis), "Hypothesis should be DSHypothesis (and not None)"
+            if not isinstance(exp.hypothesis, DSHypothesis):
+                raise TypeError("Hypothesis should be DSHypothesis (and not None)")
            if exp.hypothesis.component == component and fb:
                return True
        return False
@@ -182,7 +182,7 @@ class ExpGen2Hypothesis(DSProposalV2ExpGen):

        success_fb_list = list(set(trace_fbs))
        logger.info(
-            f"Merge Hypothesis: select {len(success_fb_list)} from {len(trace_fbs)} SOTA experiments found in {len(leaves)} traces"
+            f"Merge Hypothesis: select {len(success_fb_list)} from {len(trace_fbs)} SOTA experiments found in {len(leaves)} traces"  # nosec B608 — not SQL, Bandit false positive on "select" in log message
        )

        if len(success_fb_list) > 0:
@@ -377,7 +377,8 @@ class ExpGen2TraceAndMergeV2(ExpGen):
            if DS_RD_SETTING.enable_multi_version_exp_gen:
                exp_gen_version_list = DS_RD_SETTING.exp_gen_version_list.split(",")
                for version in exp_gen_version_list:
-                    assert version in ["v3", "v2", "v1"]
+                    if version not in ["v3", "v2", "v1"]:
+                        raise ValueError(f"version must be 'v1', 'v2', or 'v3', got {version!r}")

                if len(trace.hist) == 0:
                    # set the proposal version for the first sub-trace
@@ -339,7 +339,8 @@ class DSProposalV1ExpGen(ExpGen):
            eda_output = sota_exp.experiment_workspace.file_dict.get("EDA.md", None)
        scenario_desc = trace.scen.get_scenario_all_desc(eda_output=eda_output)

-        assert sota_exp is not None, "SOTA experiment is not provided."
+        if sota_exp is None:
+            raise ValueError("SOTA experiment is not provided.")
        last_exp = trace.last_exp()
        # exp_and_feedback = trace.hist[-1]
        # last_exp = exp_and_feedback[0]
@@ -445,8 +446,10 @@ class DSProposalV1ExpGen(ExpGen):
                        json_target_type=dict[str, dict[str, str | dict] | str],
                    )
                )
-                assert "hypothesis_proposal" in resp_dict, "Hypothesis proposal not provided."
-                assert "task_design" in resp_dict, "Task design not provided."
+                if "hypothesis_proposal" not in resp_dict:
+                    raise ValueError("Hypothesis proposal not provided.")
+                if "task_design" not in resp_dict:
+                    raise ValueError("Task design not provided.")
                task_class = component_info["task_class"]
                hypothesis_proposal = resp_dict.get("hypothesis_proposal", {})
                hypothesis = DSHypothesis(
@@ -1149,8 +1152,10 @@ You help users retrieve relevant knowledge from community discussions and public
        )

        response_dict = json.loads(response)
-        assert response_dict.get("component") in HypothesisComponent.__members__, f"Invalid component"
-        assert response_dict.get("hypothesis") is not None, f"Invalid hypothesis"
+        if response_dict.get("component") not in HypothesisComponent.__members__:
+            raise ValueError(f"Invalid component: {response_dict.get('component')}")
+        if response_dict.get("hypothesis") is None:
+            raise ValueError("Invalid hypothesis")
        return response_dict

    # END: for support llm-based hypothesis selection  -----
@@ -1253,7 +1258,8 @@ You help users retrieve relevant knowledge from community discussions and public
            description=task_desc,
        )

-        assert isinstance(task, PipelineTask), f"Task {task_name} is not a PipelineTask, got {type(task)}"
+        if not isinstance(task, PipelineTask):
+            raise TypeError(f"Task {task_name} is not a PipelineTask, got {type(task)}")
        # only for llm with response schema.(TODO: support for non-schema llm?)
        # If the LLM provides a "packages" field (list[str]), compute runtime environment now and cache it for subsequent prompts in later loops.
        if isinstance(task_dict, dict) and "packages" in task_dict and isinstance(task_dict["packages"], list):
@@ -1,3 +1,4 @@
+import ast
 import json
 import os
 import pickle
@@ -292,7 +293,7 @@ class ValidationSelector(SOTAexpSelector):
        Sorts all valid experiments by score and returns the top N.
        """

-        mock_folder = f"/tmp/mock/{self.competition}"
+        mock_folder = f"/tmp/mock/{self.competition}"  # nosec B108 — Docker volume mount point derived from internal competition name

        try:
            data_py_code, grade_py_code = self._prepare_validation_scripts(
@@ -539,7 +540,7 @@ def process_experiment(

        # Run main script
        env = get_ds_env(
-            extra_volumes={f"/tmp/mock/{competition}/{input_folder}": input_folder},
+            extra_volumes={f"/tmp/mock/{competition}/{input_folder}": input_folder},  # nosec B108 — Docker volume mount point derived from internal competition name
            running_timeout_period=DS_RD_SETTING.full_timeout,
        )
        result = ws.run(env=env, entry="python main.py")
@@ -587,8 +588,8 @@ def _parsing_score(grade_stdout: str) -> Optional[float]:
        except:
            pass
        try:
-            # Priority 2: Eval dict
-            return float(eval(json_str)["score"])
+            # Priority 2: safe literal eval for Python-style dicts
+            return float(ast.literal_eval(json_str)["score"])
        except:
            pass
        try:
@@ -35,10 +35,11 @@ def select(X: pd.DataFrame) -> pd.DataFrame:
 class KGModelFeatureSelectionCoder(Developer[KGModelExperiment]):
    def develop(self, exp: KGModelExperiment) -> KGModelExperiment:
        target_model_type = exp.sub_tasks[0].model_type
-        assert target_model_type in KG_SELECT_MAPPING
+        if target_model_type not in KG_SELECT_MAPPING:
+            raise ValueError(f"target_model_type {target_model_type} not in KG_SELECT_MAPPING")
        if len(exp.experiment_workspace.data_description) == 1:
            code = (
-                Environment(undefined=StrictUndefined)
+                Environment(undefined=StrictUndefined)  # nosec B701 — renders Python code templates, not HTML; autoescape would corrupt code
                .from_string(DEFAULT_SELECTION_CODE)
                .render(feature_index_list=None)
            )
@@ -62,7 +63,7 @@ class KGModelFeatureSelectionCoder(Developer[KGModelExperiment]):
            chosen_index_to_list_index = [i - 1 for i in chosen_index]

            code = (
-                Environment(undefined=StrictUndefined)
+                Environment(undefined=StrictUndefined)  # nosec B701 — renders Python code templates, not HTML; autoescape would corrupt code
                .from_string(DEFAULT_SELECTION_CODE)
                .render(feature_index_list=chosen_index_to_list_index)
            )
@@ -165,7 +165,8 @@ class KGScenario(Scenario):
        return data_info

    def output_format(self, tag=None) -> str:
-        assert tag in [None, "feature", "model"]
+        if tag not in [None, "feature", "model"]:
+            raise ValueError(f"tag must be None, 'feature', or 'model', got {tag!r}")
        feature_output_format = f"""The feature code should output following the format:
 {T(".prompts:kg_feature_output_format").r()}"""
        model_output_format = f"""The model code should output following the format:\n""" + T(
@@ -180,7 +181,8 @@ class KGScenario(Scenario):
            return model_output_format

    def interface(self, tag=None) -> str:
-        assert tag in [None, "feature", "XGBoost", "RandomForest", "LightGBM", "NN"]
+        if tag not in [None, "feature", "XGBoost", "RandomForest", "LightGBM", "NN"]:
+            raise ValueError(f"tag must be None, 'feature', 'XGBoost', 'RandomForest', 'LightGBM', or 'NN', got {tag!r}")
        feature_interface = f"""The feature code should follow the interface:
 {T(".prompts:kg_feature_interface").r()}"""
        if tag == "feature":
@@ -195,7 +197,8 @@ class KGScenario(Scenario):
            return model_interface

    def simulator(self, tag=None) -> str:
-        assert tag in [None, "feature", "model"]
+        if tag not in [None, "feature", "model"]:
+            raise ValueError(f"tag must be None, 'feature', or 'model', got {tag!r}")

        kg_feature_simulator = (
            "The feature code will be sent to the simulator:\n" + T(".prompts:kg_feature_simulator").r()
@@ -79,12 +79,12 @@ def preprocess_script():
    This method applies the preprocessing steps to the training, validation, and test datasets.
    """
    if os.path.exists("/kaggle/input/X_train.pkl"):
-        X_train = pd.read_pickle("/kaggle/input/X_train.pkl")
-        X_valid = pd.read_pickle("/kaggle/input/X_valid.pkl")
-        y_train = pd.read_pickle("/kaggle/input/y_train.pkl")
-        y_valid = pd.read_pickle("/kaggle/input/y_valid.pkl")
-        X_test = pd.read_pickle("/kaggle/input/X_test.pkl")
-        others = pd.read_pickle("/kaggle/input/others.pkl")
+        X_train = pd.read_pickle("/kaggle/input/X_train.pkl")  # nosec B301 — trusted Kaggle input
+        X_valid = pd.read_pickle("/kaggle/input/X_valid.pkl")  # nosec B301
+        y_train = pd.read_pickle("/kaggle/input/y_train.pkl")  # nosec B301
+        y_valid = pd.read_pickle("/kaggle/input/y_valid.pkl")  # nosec B301
+        X_test = pd.read_pickle("/kaggle/input/X_test.pkl")  # nosec B301
+        others = pd.read_pickle("/kaggle/input/others.pkl")  # nosec B301
        y_train = pd.Series(y_train).reset_index(drop=True)
        y_valid = pd.Series(y_valid).reset_index(drop=True)

@@ -87,7 +87,12 @@ def crawl_descriptions(
        content = e.get_attribute("innerHTML")
        contents.append(content)

-    assert len(subtitles) == len(contents) + 1 and subtitles[-1] == "Citation"
+    if not (len(subtitles) == len(contents) + 1 and subtitles[-1] == "Citation"):
+        raise AssertionError(
+            f"Expected len(contents)+1 == len(subtitles) and last subtitle == 'Citation', "
+            f"got len(subtitles)={len(subtitles)}, len(contents)={len(contents)}, "
+            f"last subtitle={subtitles[-1]!r}"
+        )
    for i in range(len(subtitles) - 1):
        descriptions[subtitles[i]] = contents[i]

@@ -307,7 +307,8 @@ class KGHypothesisGen(FactorAndModelHypothesisGen):
 class KGHypothesis2Experiment(FactorAndModelHypothesis2Experiment):
    def prepare_context(self, hypothesis: Hypothesis, trace: Trace) -> Tuple[dict, bool]:
        scenario = trace.scen.get_scenario_all_desc(filtered_tag="hypothesis_and_experiment")
-        assert isinstance(hypothesis, KGHypothesis)
+        if not isinstance(hypothesis, KGHypothesis):
+            raise TypeError("hypothesis must be an instance of KGHypothesis")
        experiment_output_format = (
            T("scenarios.kaggle.prompts:feature_experiment_output_format").r()
            if hypothesis.action in [KG_ACTION_FEATURE_ENGINEERING, KG_ACTION_FEATURE_PROCESSING]
@@ -1,4 +1,6 @@
+import sys
 import os
+import logging
 from pathlib import Path
 """
 Qlib Factor Runner - Executes factor backtests in Docker.
@@ -29,6 +31,83 @@ from rdagent.scenarios.qlib.experiment.model_experiment import QlibModelExperime
 DIRNAME = Path(__file__).absolute().resolve().parent
 DIRNAME_local = Path.cwd()

+
+def _shift_daily_constant_factor_if_needed(factor_col: "pd.Series", factor_name: str) -> "pd.Series":
+    """Detect and fix look-ahead bias in daily-constant factors.
+
+    A factor is "daily-constant" when every minute bar within the same calendar
+    day carries an identical value. This happens when LLM code computes a daily
+    aggregate (e.g. today's log return) and forward-fills it across all intraday
+    bars without shifting — meaning the end-of-day value is visible at 00:00.
+
+    Fix: shift by one trading day so that the value assigned to day T is the
+    aggregate computed from day T-1, eliminating the forward-looking information.
+    """
+    import numpy as np
+
+    try:
+        notnull = factor_col.dropna()
+        if len(notnull) < 200:
+            return factor_col
+
+        datetimes = notnull.index.get_level_values("datetime")
+        dates = datetimes.normalize()
+
+        # Sample up to 50 random days and check intra-day uniqueness
+        unique_dates = pd.Series(dates.unique())
+        sample_dates = unique_dates.sample(min(50, len(unique_dates)), random_state=42)
+
+        daily_unique_counts = []
+        for d in sample_dates:
+            mask = dates == d
+            vals = notnull.values[mask]
+            if len(vals) > 1:
+                daily_unique_counts.append(len(np.unique(vals[~np.isnan(vals)])))
+
+        if not daily_unique_counts:
+            return factor_col
+
+        # If >90% of sampled days have exactly 1 unique value → daily-constant
+        fraction_constant = sum(1 for c in daily_unique_counts if c == 1) / len(daily_unique_counts)
+        if fraction_constant < 0.90:
+            return factor_col  # Intraday factor — no shift needed
+
+        logger.warning(
+            f"[LookAheadFix] Factor '{factor_name}' is daily-constant "
+            f"({fraction_constant:.0%} of days). Applying 1-day shift to remove look-ahead bias."
+        )
+
+        # Shift: for each instrument, map daily values forward by 1 trading day
+        instruments = factor_col.index.get_level_values("instrument").unique()
+        shifted_parts = []
+        for inst in instruments:
+            inst_series = factor_col.xs(inst, level="instrument")
+            # Get one value per calendar day (the first non-null bar)
+            inst_dt = inst_series.index.normalize()
+            daily_vals = inst_series.groupby(inst_dt).first()
+            # Shift by 1 day
+            daily_vals_shifted = daily_vals.shift(1)
+            # Forward-fill back to minute bars
+            minute_idx = inst_series.index
+            minute_dates = minute_idx.normalize()
+            shifted_minute = minute_dates.map(daily_vals_shifted)
+            shifted_s = pd.Series(
+                shifted_minute.values,
+                index=pd.MultiIndex.from_arrays(
+                    [inst_series.index, [inst] * len(inst_series)],
+                    names=["datetime", "instrument"],
+                ),
+                name=factor_col.name,
+            )
+            shifted_parts.append(shifted_s)
+
+        return pd.concat(shifted_parts).sort_index()
+
+    except Exception as e:
+        logger.debug(f"[LookAheadFix] Could not apply daily shift for '{factor_name}': {e}")
+        return factor_col
+
+
 # TODO: supporting multiprocessing and keep previous results


@@ -391,8 +470,19 @@ class QlibFactorRunner(CachedRunner[QlibFactorExperiment]):
        import numpy as np

        try:
-            # Get workspace path
-            workspace_path = exp.experiment_workspace.workspace_path
+            # Get workspace path — factor code and result.h5 live in sub_workspace_list[0],
+            # not in experiment_workspace (which is the Qlib template workspace).
+            workspace_path = None
+            if exp.sub_workspace_list:
+                for ws in exp.sub_workspace_list:
+                    if ws is not None and hasattr(ws, 'workspace_path'):
+                        candidate = ws.workspace_path / "result.h5"
+                        if candidate.exists():
+                            workspace_path = ws.workspace_path
+                            break
+            if workspace_path is None:
+                # Fallback to experiment_workspace
+                workspace_path = exp.experiment_workspace.workspace_path
            if workspace_path is None:
                return None

@@ -409,6 +499,12 @@ class QlibFactorRunner(CachedRunner[QlibFactorExperiment]):
            factor_col = factor_values.iloc[:, 0]
            factor_name = factor_values.columns[0]

+            # Detect and fix look-ahead bias in daily-constant factors.
+            # If a factor has the same value for all minute bars within each calendar day
+            # it was computed from same-day data (e.g. today's close return at 00:00).
+            # Fix: shift by 1 trading day so value at day T = aggregate of day T-1.
+            factor_col = _shift_daily_constant_factor_if_needed(factor_col, factor_name)
+
            # Load source data for forward returns
            data_path = (
                Path(__file__).parent.parent.parent.parent.parent
@@ -587,10 +683,12 @@ class QlibFactorRunner(CachedRunner[QlibFactorExperiment]):
            from pathlib import Path
            from rdagent.components.backtesting import ResultsDatabase

-            # Get factor name from hypothesis
+            # Get factor name: prefer hypothesis, fallback to result Series 'factor_name' key
            factor_name = "unknown"
            if hasattr(exp, 'hypothesis') and exp.hypothesis is not None:
                factor_name = getattr(exp.hypothesis, 'hypothesis', 'unknown')
+            if factor_name == 'unknown' and isinstance(result, pd.Series) and 'factor_name' in result.index:
+                factor_name = str(result['factor_name'])

            # Check if already rejected by protection
            if getattr(exp, 'rejected_by_protection', False):
@@ -824,41 +922,74 @@ class QlibFactorRunner(CachedRunner[QlibFactorExperiment]):
        """
        Save factor time-series values as parquet for strategy building.

-        This is essential for walk-forward validation and strategy combination.
-
-        Parameters
-        ----------
-        factor_name : str
-            Name of the factor
-        exp : QlibFactorExperiment
-            The experiment with factor values
+        Reruns the factor code on the FULL 6-year dataset so the parquet covers
+        the complete backtest range (not just the debug 2024 subset).
        """
        import os as _os
+        import subprocess
+        import shutil
+        import tempfile

        try:
-            # Get workspace path
-            workspace_path = exp.experiment_workspace.workspace_path
+            # factor.py lives in sub_workspace_list[0], not experiment_workspace
+            workspace_path = None
+            if exp.sub_workspace_list:
+                for ws in exp.sub_workspace_list:
+                    if ws is not None and hasattr(ws, 'workspace_path'):
+                        fp = ws.workspace_path / "factor.py"
+                        if fp.exists():
+                            workspace_path = ws.workspace_path
+                            break
+            if workspace_path is None:
+                workspace_path = exp.experiment_workspace.workspace_path
            if workspace_path is None:
                return

-            result_h5 = workspace_path / "result.h5"
-            if not result_h5.exists():
+            factor_py = workspace_path / "factor.py"
+            if not factor_py.exists():
                return

-            # Read factor values
+            project_root = Path(__file__).parent.parent.parent.parent.parent
+            full_data = (
+                project_root
+                / "git_ignore_folder"
+                / "factor_implementation_source_data"
+                / "intraday_pv.h5"
+            )
+            if not full_data.exists():
+                return
+
+            # Run factor code on full data in a temp workspace
            import pandas as pd
-            df = pd.read_hdf(str(result_h5), key="data")
+            with tempfile.TemporaryDirectory(prefix="predix_fullval_") as tmp_dir:
+                tmp = Path(tmp_dir)
+                shutil.copy(str(factor_py), str(tmp / "factor.py"))
+                shutil.copy(str(full_data), str(tmp / "intraday_pv.h5"))
+
+                ret = subprocess.run(
+                    ["sys.executable", "factor.py"],
+                    cwd=str(tmp),
+                    capture_output=True,
+                    timeout=300,
+                )
+                if ret.returncode != 0:
+                    # Fall back to debug-data result if full-data run fails
+                    result_h5 = workspace_path / "result.h5"
+                    if not result_h5.exists():
+                        return
+                    df = pd.read_hdf(str(result_h5), key="data")
+                else:
+                    result_h5_full = tmp / "result.h5"
+                    if not result_h5_full.exists():
+                        return
+                    df = pd.read_hdf(str(result_h5_full), key="data")
+
            if df is None or df.empty:
                return

-            # Get the factor series (first column)
            series = df.iloc[:, 0]
            series.name = factor_name

-            # Save to results/factors/values/
-            project_root = Path(__file__).parent.parent.parent.parent.parent
-
-            # Parallel run isolation
            parallel_run_id = _os.getenv("PARALLEL_RUN_ID", "0")
            if parallel_run_id != "0":
                values_dir = project_root / "results" / "runs" / f"run{parallel_run_id}" / "factors" / "values"
@@ -866,17 +997,12 @@ class QlibFactorRunner(CachedRunner[QlibFactorExperiment]):
                values_dir = project_root / "results" / "factors" / "values"

            values_dir.mkdir(parents=True, exist_ok=True)
-
-            # Safe filename
            safe_name = factor_name.replace("/", "_").replace("\\", "_").replace(" ", "_")[:100]
            parquet_path = values_dir / f"{safe_name}.parquet"
+            series.to_frame().to_parquet(str(parquet_path))

-            # Save as parquet (with datetime index)
-            series.to_parquet(str(parquet_path))
-
-        except Exception as e:
-            # Don't let factor value saving break the main workflow
-            pass
+        except Exception:
+            logging.debug("Error in save_factor_values_to_parquet", exc_info=True)

    def _log_result_warnings(self, factor_name: str, result, metrics: dict) -> None:
        """
@@ -241,6 +241,7 @@ class StrategyBuilder:
                if data.get("status") == "success" and data.get("ic") is not None:
                    factors.append(data)
            except Exception:
+                logger.warning("Failed to load factor file %s", f, exc_info=True)
                continue

        # Sort by absolute IC
@@ -30,7 +30,8 @@ def _build_execute_calls(exp: QlibFactorExperiment, base_feature_workspaces: lis
    execute_calls = []

    if exp.sub_tasks:
-        assert isinstance(exp.prop_dev_feedback, CoSTEERMultiFeedback)
+        if not isinstance(exp.prop_dev_feedback, CoSTEERMultiFeedback):
+            raise TypeError("exp.prop_dev_feedback must be of type CoSTEERMultiFeedback")
        execute_calls.extend(
            (implementation.execute, ("All",))
            for implementation, feedback in zip(exp.sub_workspace_list, exp.prop_dev_feedback)
@@ -23,14 +23,25 @@ $low: low price at 1-minute bar.
 $volume: volume at 1-minute bar (tick volume for FX).

 ## Important Notes for 1min Data
- 96 bars = 1 trading day (24 hours for FX)
+- 1 bar = 1 minute (confirmed)
 - 16 bars = 16 minutes
- 4 bars = 4 minutes
- 1 bar = 1 minute
+- 60 bars = 1 hour
+- ~1440 bars = 1 full trading day (FX trades nearly 24h, Mon 00:00 - Fri 22:00 UTC approx.)
+- Typical bars per calendar day: ~1200-1440 (varies by weekday, holidays have fewer)
+- Do NOT assume 96 bars/day — the actual count depends on the date
 - Data range: 2020-01-01 to 2026-03-20
 - Instrument: EURUSD
 - Timezone: UTC

+## IMPORTANT: Bars per Day Correction
+The dataset has approximately 1440 bars per full trading day (1 bar = 1 minute, ~24h of FX trading).
+Some older documentation incorrectly stated "96 bars = 1 day" — this is WRONG. Always use:
+- 60 bars = 1 hour
+- 480 bars = 8 hours (London session 08:00-16:00 UTC)
+- 180 bars = 3 hours (London/NY overlap 13:00-16:00 UTC)
+Use datetime hour filtering (e.g., `df[df.index.get_level_values('datetime').hour.between(8, 15)]`)
+to select session bars — do NOT use bar-count offsets to define sessions.
+
 ## Session Times (UTC)
 - Asian: 00:00-08:00 UTC (low volatility)
 - London: 08:00-16:00 UTC (high volatility)
@@ -104,7 +104,7 @@ qlib_factor_strategy: |-
     result_df.columns = ['daily_volume_price_divergence']
     ```

-  4. **Process ALL data — do not filter dates**: The source HDF5 contains data from 2020-01-01 to 2026-03-20. Do NOT filter to a single year. If your output has only 314 entries (one year of daily data), the factor will be rejected. Expected output: ~1500+ daily entries for 2020-2026.
+  4. **Process ALL data — do not filter dates**: The source HDF5 contains data from 2020-01-01 to 2026-03-20 (development runs may use a 2024-only debug dataset with ~300 entries, which is acceptable). Do NOT filter to a single year in your code. Write your code to process whatever date range is available in the HDF5 file — do not hardcode date filters. Expected output for production data: ~1500+ daily entries for 2020-2026. Expected output for debug data: ~300 daily entries for 2024. Both are valid.

  5. **Use `transform()` instead of `apply()` for per-group calculations**: `transform()` preserves the original index while `apply()` may reduce the number of rows unexpectedly:
     ```python
@@ -121,6 +121,35 @@ qlib_factor_strategy: |-
     assert result_df.index.names == ['datetime', 'instrument'], f"Index names must be ['datetime', 'instrument'], got {result_df.index.names}"
     ```

+  7. **NEVER use same-day aggregations as the factor value — always shift by 1 day**: If your factor computes a daily aggregate (e.g. daily close return, daily OHLC range, daily volume), that aggregate is only known at end-of-day. Using it at the start of the same day is look-ahead bias. You MUST shift the daily aggregate by 1 day before forward-filling to minute bars:
+     ```python
+     # WRONG: look-ahead bias! Today's close return is not known at 00:00
+     daily_ret = df['$close'].groupby(level='instrument').resample('1D', level='datetime').last().pct_change()
+     result_df['my_factor'] = daily_ret.groupby(level='instrument').transform(lambda x: x.reindex(df.index.get_level_values('datetime'), method='ffill'))
+
+     # CORRECT: shift by 1 trading day so factor value at day T = aggregate of day T-1
+     daily_close = df.groupby([df.index.get_level_values('datetime').normalize(), df.index.get_level_values('instrument')])['$close'].last()
+     daily_close.index.names = ['date', 'instrument']
+     daily_ret = daily_close.groupby(level='instrument').pct_change().shift(1)  # <-- shift(1) is MANDATORY
+     # then map back to minute bars via ffill
+     ```
+     This rule applies to ALL daily aggregations: returns, OHLC stats, volume, momentum, slopes, etc.
+     **Session-based aggregations (London, NY, Asian session returns) are also daily aggregations** — the London
+     session (08:00-16:00 UTC) ends at 16:00, so its return must be shifted by 1 day before use.
+     Intraday rolling factors (e.g. 30-min rolling std computed at bar t using only bars t-N..t-1) do NOT need this shift.
+
+  8. **PREFER pure intraday rolling factors**: Factors that use only a trailing window of recent bars (e.g.
+     rolling(30).mean() of returns, RSI(14), Bollinger Band z-score) have NO look-ahead risk and vary every
+     minute. These are the best candidates for short-horizon (60-180 bar) prediction. Examples:
+     - Rolling 15-min / 30-min / 60-min return momentum (15, 30, 60 bars respectively)
+     - Rolling volatility (std of returns over 20-60 bars)
+     - Distance of close from N-bar moving average (z-score)
+     - RSI or similar oscillators computed on 1-min bars
+     - VWAP deviation (requires volume — use $volume column)
+     Always use `.shift(1)` on the lagged window (e.g. `rolling(N).mean().shift(1)`) to avoid using the
+     current bar's own price in its own feature value.
+     NOTE: 1 bar = 1 minute. The data has ~1440 bars per full trading day. Do NOT use 96 as a day proxy.
+
 qlib_factor_output_format: |-
  Your output should be a pandas dataframe similar to the following example information:
  <class 'pandas.core.frame.DataFrame'>
@@ -56,7 +56,8 @@ class QlibQuantScenario(Scenario):
        )

    def background(self, tag=None) -> str:
-        assert tag in [None, "factor", "model"]
+        if tag not in [None, "factor", "model"]:
+            raise ValueError(f"tag must be None, 'factor', or 'model', got {tag!r}")
        quant_background = "The background of the scenario is as follows:\n" + T(".prompts:qlib_quant_background").r(
            runtime_environment=self.get_runtime_environment(),
        )
@@ -83,7 +84,8 @@ class QlibQuantScenario(Scenario):
        return self._source_data

    def output_format(self, tag=None) -> str:
-        assert tag in [None, "factor", "model"]
+        if tag not in [None, "factor", "model"]:
+            raise ValueError(f"tag must be None, 'factor', or 'model', got {tag!r}")
        factor_output_format = (
            "The factor code should output the following format:\n" + T(".prompts:qlib_factor_output_format").r()
        )
@@ -99,7 +101,8 @@ class QlibQuantScenario(Scenario):
            return model_output_format

    def interface(self, tag=None) -> str:
-        assert tag in [None, "factor", "model"]
+        if tag not in [None, "factor", "model"]:
+            raise ValueError(f"tag must be None, 'factor', or 'model', got {tag!r}")
        factor_interface = (
            "The factor code should be written in the following interface:\n" + T(".prompts:qlib_factor_interface").r()
        )
@@ -115,7 +118,8 @@ class QlibQuantScenario(Scenario):
            return model_interface

    def simulator(self, tag=None) -> str:
-        assert tag in [None, "factor", "model"]
+        if tag not in [None, "factor", "model"]:
+            raise ValueError(f"tag must be None, 'factor', or 'model', got {tag!r}")
        factor_simulator = "The factor code will be sent to the simulator:\n" + T(".prompts:qlib_factor_simulator").r()
        model_simulator = "The model code will be sent to the simulator:\n" + T(".prompts:qlib_model_simulator").r()

@@ -185,7 +189,8 @@ class QlibQuantScenario(Scenario):
            return common_description(action) + interface(action) + output(action) + simulator(action)

    def get_runtime_environment(self, tag: str = None) -> str:
-        assert tag in [None, "factor", "model"]
+        if tag not in [None, "factor", "model"]:
+            raise ValueError(f"tag must be None, 'factor', or 'model', got {tag!r}")

        if tag is None or tag == "factor":
            # Use factor env to get the runtime environment
@@ -4,7 +4,7 @@ import shutil
 from pathlib import Path

 import pandas as pd
-from jinja2 import Environment, StrictUndefined
+from jinja2 import Environment, StrictUndefined, select_autoescape

 from rdagent.components.coder.factor_coder.config import FACTOR_COSTEER_SETTINGS
 from rdagent.utils.env import QTDockerEnv
@@ -21,14 +21,16 @@ def generate_data_folder_from_qlib():
        entry=f"python generate.py",
    )

-    assert (Path(__file__).parent / "factor_data_template" / "intraday_pv_all.h5").exists(), (
-        "intraday_pv_all.h5 is not generated. It means rdagent/scenarios/qlib/experiment/factor_data_template/generate.py is not executed correctly. Please check the log: \n"
-        + execute_log
-    )
-    assert (Path(__file__).parent / "factor_data_template" / "intraday_pv_debug.h5").exists(), (
-        "intraday_pv_debug.h5 is not generated. It means rdagent/scenarios/qlib/experiment/factor_data_template/generate.py is not executed correctly. Please check the log: \n"
-        + execute_log
-    )
+    if not (Path(__file__).parent / "factor_data_template" / "intraday_pv_all.h5").exists():
+        raise FileNotFoundError(
+            "intraday_pv_all.h5 is not generated. It means rdagent/scenarios/qlib/experiment/factor_data_template/generate.py is not executed correctly. Please check the log: \n"
+            + execute_log
+        )
+    if not (Path(__file__).parent / "factor_data_template" / "intraday_pv_debug.h5").exists():
+        raise FileNotFoundError(
+            "intraday_pv_debug.h5 is not generated. It means rdagent/scenarios/qlib/experiment/factor_data_template/generate.py is not executed correctly. Please check the log: \n"
+            + execute_log
+        )

    Path(FACTOR_COSTEER_SETTINGS.data_folder).mkdir(parents=True, exist_ok=True)
    shutil.copy(
@@ -67,7 +69,7 @@ def get_file_desc(p: Path, variable_list=[]) -> str:
    """
    p = Path(p)

-    JJ_TPL = Environment(undefined=StrictUndefined).from_string("""
+    JJ_TPL = Environment(undefined=StrictUndefined, autoescape=select_autoescape()).from_string("""
 # {{file_name}}

 ## File Type
@@ -1,4 +1,6 @@
+import logging
 import json
+import os
 from typing import List, Tuple

 from rdagent.components.coder.factor_coder.factor import FactorExperiment, FactorTask
@@ -9,6 +11,47 @@ from rdagent.scenarios.qlib.experiment.model_experiment import QlibModelExperime
 from rdagent.scenarios.qlib.experiment.quant_experiment import QlibQuantScenario
 from rdagent.utils.agent.tpl import T

+
+def _build_compressed_history(trace: Trace, max_history: int) -> str:
+    """Return hypothesis_and_feedback string with only `max_history` entries.
+
+    Older entries beyond the last 2 are compressed to one bullet line each.
+    """
+    if len(trace.hist) == 0:
+        return "No previous hypothesis and feedback available since it's the first round."
+
+    FULL_DETAIL = 2
+    old_hist = trace.hist[:-FULL_DETAIL] if len(trace.hist) > FULL_DETAIL else []
+    recent_hist = trace.hist[-FULL_DETAIL:] if len(trace.hist) > FULL_DETAIL else trace.hist
+
+    parts = []
+    if old_hist:
+        lines = ["## Earlier experiments (summarized):"]
+        for exp, fb in old_hist:
+            names = []
+            for task in exp.sub_tasks:
+                if task is not None and hasattr(task, "factor_name"):
+                    names.append(task.factor_name)
+                elif task is not None and hasattr(task, "model_type"):
+                    names.append(getattr(task, "model_type", "model"))
+            ic_str = ""
+            try:
+                if exp.result is not None and "IC" in exp.result.index:
+                    ic_str = f" IC={exp.result.loc['IC']:.4f}"
+            except Exception:
+                logging.debug("Exception caught", exc_info=True)
+            decision = "PASS" if fb.decision else "FAIL"
+            obs = (fb.observations or "")[:120].replace("\n", " ")
+            lines.append(f"- [{decision}]{ic_str} {', '.join(names) or 'unknown'}: {obs}")
+        parts.append("\n".join(lines))
+
+    if recent_hist:
+        rt = Trace(trace.scen)
+        rt.hist = recent_hist
+        parts.append(T("scenarios.qlib.prompts:hypothesis_and_feedback").r(trace=rt))
+
+    return "\n\n".join(parts)
+
 QlibFactorHypothesis = Hypothesis


@@ -17,13 +60,10 @@ class QlibFactorHypothesisGen(FactorHypothesisGen):
        super().__init__(scen)

    def prepare_context(self, trace: Trace) -> Tuple[dict, bool]:
-        hypothesis_and_feedback = (
-            T("scenarios.qlib.prompts:hypothesis_and_feedback").r(
-                trace=trace,
-            )
-            if len(trace.hist) > 0
-            else "No previous hypothesis and feedback available since it's the first round."
-        )
+        max_h = int(os.environ.get("QLIB_QUANT_MAX_FACTOR_HISTORY", "20"))
+        limited = Trace(trace.scen)
+        limited.hist = trace.hist[-max_h:] if len(trace.hist) > max_h else trace.hist
+        hypothesis_and_feedback = _build_compressed_history(limited, max_h)
        last_hypothesis_and_feedback = (
            T("scenarios.qlib.prompts:last_hypothesis_and_feedback").r(
                experiment=trace.hist[-1][0], feedback=trace.hist[-1][1]
@@ -70,15 +110,15 @@ class QlibFactorHypothesis2Experiment(FactorHypothesis2Experiment):
        if len(trace.hist) == 0:
            hypothesis_and_feedback = "No previous hypothesis and feedback available since it's the first round."
        else:
+            max_h = int(os.environ.get("QLIB_QUANT_MAX_FACTOR_HISTORY", "20"))
+            factor_hist = [
+                e for e in trace.hist
+                if not hasattr(e[0].hypothesis, "action") or e[0].hypothesis.action == "factor"
+            ][-max_h:]
            specific_trace = Trace(trace.scen)
-            for i in range(len(trace.hist) - 1, -1, -1):
-                if not hasattr(trace.hist[i][0].hypothesis, "action") or trace.hist[i][0].hypothesis.action == "factor":
-                    specific_trace.hist.insert(0, trace.hist[i])
-            if len(specific_trace.hist) > 0:
-                specific_trace.hist.reverse()
-                hypothesis_and_feedback = T("scenarios.qlib.prompts:hypothesis_and_feedback").r(
-                    trace=specific_trace,
-                )
+            specific_trace.hist = factor_hist
+            if specific_trace.hist:
+                hypothesis_and_feedback = _build_compressed_history(specific_trace, max_h)
            else:
                hypothesis_and_feedback = "No previous hypothesis and feedback available."

@@ -1,3 +1,4 @@
+import logging
 import json
 import os
 import random
@@ -152,9 +153,41 @@ class QlibQuantHypothesisGen(FactorAndModelHypothesisGen):
                        factor_inserted = True
            if len(specific_trace.hist) > 0:
                specific_trace.hist.reverse()
-                hypothesis_and_feedback = T("scenarios.qlib.prompts:hypothesis_and_feedback").r(
-                    trace=specific_trace,
-                )
+                # Keep only the 2 most recent experiments in full detail; compress older ones
+                # to brief bullet points to stay within the LLM context window.
+                FULL_DETAIL_COUNT = 2
+                old_hist = specific_trace.hist[:-FULL_DETAIL_COUNT] if len(specific_trace.hist) > FULL_DETAIL_COUNT else []
+                recent_hist = specific_trace.hist[-FULL_DETAIL_COUNT:] if len(specific_trace.hist) > FULL_DETAIL_COUNT else specific_trace.hist
+
+                parts = []
+                if old_hist:
+                    summary_lines = ["## Earlier experiments (summarized):"]
+                    for exp, fb in old_hist:
+                        factor_names = []
+                        for task in exp.sub_tasks:
+                            if task is not None and hasattr(task, "factor_name"):
+                                factor_names.append(task.factor_name)
+                            elif task is not None and hasattr(task, "model_type"):
+                                factor_names.append(getattr(task, "model_type", "model"))
+                        names_str = ", ".join(factor_names) if factor_names else "unknown"
+                        ic_str = ""
+                        try:
+                            if exp.result is not None:
+                                ic_val = exp.result.loc["IC"] if "IC" in exp.result.index else ""
+                                ic_str = f" IC={ic_val:.4f}" if ic_val != "" else ""
+                        except Exception:
+                            logging.debug("Error getting IC", exc_info=True)
+                        decision_str = "PASS" if fb.decision else "FAIL"
+                        obs_short = (fb.observations or "")[:120].replace("\n", " ")
+                        summary_lines.append(f"- [{decision_str}]{ic_str} {names_str}: {obs_short}")
+                    parts.append("\n".join(summary_lines))
+
+                if recent_hist:
+                    recent_trace = Trace(specific_trace.scen)
+                    recent_trace.hist = recent_hist
+                    parts.append(T("scenarios.qlib.prompts:hypothesis_and_feedback").r(trace=recent_trace))
+
+                hypothesis_and_feedback = "\n\n".join(parts)
            else:
                hypothesis_and_feedback = "No previous hypothesis and feedback available."

@@ -77,6 +77,7 @@ def count_valid_factors() -> int:
            if data.get("status") == "success" and data.get("ic") is not None:
                count += 1
        except Exception:
+            logger.warning("Failed to load factor file %s", json_file, exc_info=True)
            continue

    return count
@@ -71,7 +71,7 @@ def submit_for_grading(grading_url: str, model_path: str) -> dict | None:
 def main():
    MODEL_PATH = os.environ.get("MODEL_PATH")
    DATA_PATH = os.environ.get("DATA_PATH")
-    OUTPUT_DIR = os.environ.get("OUTPUT_DIR", "/tmp/autorl_output")
+    OUTPUT_DIR = os.environ.get("OUTPUT_DIR", "/tmp/autorl_output")  # nosec B108 — Docker container output dir, configurable via env var
    GRADING_SERVER_URL = os.environ.get("GRADING_SERVER_URL", "")
    TRAIN_RATIO = float(os.environ.get("TRAIN_RATIO", "0.05"))
    NUM_EPOCHS = int(os.environ.get("NUM_EPOCHS", "3"))
@@ -9,7 +9,7 @@ peft>=0.18.1

 # Evaluation
 opencompass==0.5.1
-setuptools<75  # uv venv doesn't include, opencompass depends on pkg_resources
+setuptools>=78.1.1  # Security fix: GHSA-8g6x-3r52-4m6c (path traversal in PackageIndex.download, arbitrary file write/RCE)

 # Inference acceleration (optional, TRL supports 0.10.2-0.12.0)
 # Security: Version >=0.14.0 fixes CVE-2026-22807 (RCE via auto_map dynamic module loading)
@@ -9,7 +9,7 @@ from pathlib import Path
 from typing import Any

 import yaml
-from jinja2 import Environment, FunctionLoader, StrictUndefined
+from jinja2 import Environment, FunctionLoader, StrictUndefined, select_autoescape

 from rdagent.core.conf import RD_AGENT_SETTINGS
 from rdagent.log import rdagent_logger as logger
@@ -38,7 +38,8 @@ def load_content(uri: str, caller_dir: Path | None = None, ftype: str = "yaml")
        caller_dir = get_caller_dir(upshift=1)
    # Parse the URI
    path_part, *yaml_trace = uri.split(":")
-    assert len(yaml_trace) <= 1, f"Invalid uri {uri}, only one yaml trace is allowed."
+    if len(yaml_trace) > 1:
+        raise ValueError(f"Invalid uri {uri}, only one yaml trace is allowed.")
    yaml_trace = [key for yt in yaml_trace for key in yt.split(".")]

    # load file_path with priorities.
@@ -126,7 +127,7 @@ class RDAT:
        # loader=FunctionLoader(load_conent) is for supporting grammar like below.
        # `{% include "scenarios.data_science.share:component_spec.DataLoadSpec" %}`
        rendered = (
-            Environment(undefined=StrictUndefined, loader=FunctionLoader(load_content))
+            Environment(undefined=StrictUndefined, loader=FunctionLoader(load_content), autoescape=select_autoescape())
            .from_string(self.template)
            .render(**context)
            .strip("\n")
@@ -614,13 +614,14 @@ class LocalEnv(Env[ASpecificLocalConf]):
        if self.conf.extra_volumes is not None:
            for lp, rp in self.conf.extra_volumes.items():
                volumes[lp] = rp["bind"] if isinstance(rp, dict) else rp
-            cache_path = "/tmp/sample" if "/sample/" in "".join(self.conf.extra_volumes.keys()) else "/tmp/full"
+            cache_path = "/tmp/sample" if "/sample/" in "".join(self.conf.extra_volumes.keys()) else "/tmp/full"  # nosec B108 — fixed Docker volume mount point, not a user-writable temp file
            Path(cache_path).mkdir(parents=True, exist_ok=True)
            volumes[cache_path] = T("scenarios.data_science.share:scen.cache_path").r()
        for lp, rp in running_extra_volume.items():
            volumes[lp] = rp

-        assert local_path is not None, "local_path should not be None"
+        if local_path is None:
+            raise ValueError("local_path should not be None")
        volumes = normalize_volumes(volumes, local_path)

        @contextlib.contextmanager
@@ -678,7 +679,7 @@ class LocalEnv(Env[ASpecificLocalConf]):
            cwd = Path(local_path).resolve() if local_path else None
            env = {k: str(v) if isinstance(v, int) else v for k, v in env.items()}

-            process = subprocess.Popen(
+            process = subprocess.Popen(  # nosec B602 — entry is an internal command string set by LocalEnvConf, not user input
                entry,
                cwd=cwd,
                env={**os.environ, **env},
@@ -761,12 +762,15 @@ class CondaConf(LocalConf):
        to ensure bin_path is set correctly even if the conda env was just created.
        """
        conda_path_result = subprocess.run(
-            f"conda run -n {self.conda_env_name} --no-capture-output env | grep '^PATH='",
+            ["conda", "run", "-n", self.conda_env_name, "--no-capture-output", "env"],
            capture_output=True,
            text=True,
-            shell=True,
        )
-        self.bin_path = conda_path_result.stdout.strip().split("=")[1] if conda_path_result.returncode == 0 else ""
+        if conda_path_result.returncode == 0:
+            path_lines = [l for l in conda_path_result.stdout.splitlines() if l.startswith("PATH=")]
+            self.bin_path = path_lines[0].split("=", 1)[1] if path_lines else ""
+        else:
+            self.bin_path = ""


 class MLECondaConf(CondaConf):
@@ -850,24 +854,22 @@ class QlibCondaEnv(LocalEnv[QlibCondaConf]):
    def prepare(self) -> None:
        """Prepare the conda environment if not already created."""
        try:
-            envs = subprocess.run("conda env list", capture_output=True, text=True, shell=True)
+            envs = subprocess.run(["conda", "env", "list"], capture_output=True, text=True)
            if self.conf.conda_env_name not in envs.stdout:
                print(f"[yellow]Conda env '{self.conf.conda_env_name}' not found, creating...[/yellow]")
                subprocess.check_call(
-                    f"conda create -y -n {self.conf.conda_env_name} python=3.10",
-                    shell=True,
+                    ["conda", "create", "-y", "-n", self.conf.conda_env_name, "python=3.10"],
                )
                subprocess.check_call(
-                    f"conda run -n {self.conf.conda_env_name} pip install --upgrade pip cython",
-                    shell=True,
+                    ["conda", "run", "-n", self.conf.conda_env_name, "pip", "install", "--upgrade", "pip", "cython"],
                )
                subprocess.check_call(
-                    f"conda run -n {self.conf.conda_env_name} pip install git+https://github.com/microsoft/qlib.git@2fb9380b342556ddb50a4b24e4fe8655d548b2b8",
-                    shell=True,
+                    ["conda", "run", "-n", self.conf.conda_env_name, "pip", "install",
+                     "git+https://github.com/microsoft/qlib.git@2fb9380b342556ddb50a4b24e4fe8655d548b2b8"],
                )
                subprocess.check_call(
-                    f"conda run -n {self.conf.conda_env_name} pip install catboost xgboost tables torch",
-                    shell=True,
+                    ["conda", "run", "-n", self.conf.conda_env_name, "pip", "install",
+                     "catboost", "xgboost", "tables", "torch"],
                )

        except Exception as e:
@@ -888,10 +890,9 @@ def _sync_conda_cache_with_real_envs() -> None:
    """Ensure the prepared cache includes environments that already exist on disk."""
    try:
        result = subprocess.run(
-            "conda env list",
+            ["conda", "env", "list"],
            capture_output=True,
            text=True,
-            shell=True,
            check=False,
        )
    except Exception as exc:  # pragma: no cover - best-effort helper
@@ -924,14 +925,15 @@ def _prepare_conda_env(env_name: str, requirements_file: Path, python_version: s
        python_version: Python version for the environment
    """
    # 1. Create conda environment if not exists
-    result = subprocess.run(f"conda env list | grep -q '^{env_name} '", shell=True)
-    if result.returncode != 0:
+    env_list = subprocess.run(["conda", "env", "list"], capture_output=True, text=True, check=False)
+    env_exists = any(line.split()[0] == env_name for line in env_list.stdout.splitlines() if line and not line.startswith("#"))
+    if not env_exists:
        print(f"[yellow]Creating conda env '{env_name}' (Python {python_version})...[/yellow]")
-        subprocess.check_call(f"conda create -y -n {env_name} python={python_version}", shell=True)
-        subprocess.check_call(f"conda run -n {env_name} pip install --upgrade pip", shell=True)
+        subprocess.check_call(["conda", "create", "-y", "-n", env_name, f"python={python_version}"])
+        subprocess.check_call(["conda", "run", "-n", env_name, "pip", "install", "--upgrade", "pip"])

    print(f"[yellow]Installing dependencies from {requirements_file.name}...[/yellow]")
-    subprocess.check_call(f"conda run -n {env_name} pip install -r {requirements_file}", shell=True)
+    subprocess.check_call(["conda", "run", "-n", env_name, "pip", "install", "-r", str(requirements_file)])
    print(f"[green]Conda env '{env_name}' ready[/green]")

    _CONDA_ENV_PREPARED.add(env_name)
@@ -971,8 +973,8 @@ class FTCondaEnv(LocalEnv[FTCondaConf]):
            # Note: flash-attn>=2.8 is required for B200 (sm_100) support
            print("[yellow]Installing flash-attn (compiling, may take a few minutes)...[/yellow]")
            subprocess.check_call(
-                f"conda run -n {self.conf.conda_env_name} pip install 'flash-attn>=2.8' --no-build-isolation --no-cache-dir",
-                shell=True,
+                ["conda", "run", "-n", self.conf.conda_env_name, "pip", "install",
+                 "flash-attn>=2.8", "--no-build-isolation", "--no-cache-dir"],
            )

            # Re-update bin_path after prepare() in case the conda env was just created
@@ -1442,7 +1444,7 @@ class DockerEnv(Env[DockerConf]):
        if self.conf.extra_volumes is not None:
            for lp, rp in self.conf.extra_volumes.items():
                volumes[lp] = rp if isinstance(rp, dict) else {"bind": rp, "mode": self.conf.extra_volume_mode}
-            cache_path = "/tmp/sample" if "/sample/" in "".join(self.conf.extra_volumes.keys()) else "/tmp/full"
+            cache_path = "/tmp/sample" if "/sample/" in "".join(self.conf.extra_volumes.keys()) else "/tmp/full"  # nosec B108 — fixed Docker volume mount point, not a user-writable temp file
            Path(cache_path).mkdir(parents=True, exist_ok=True)
            volumes[cache_path] = {
                "bind": T("scenarios.data_science.share:scen.cache_path").r(),
@@ -1471,7 +1473,8 @@ class DockerEnv(Env[DockerConf]):
                cpu_count=self.conf.cpu_count,  # Set CPU limit
                **self._gpu_kwargs(client),
            )
-            assert container is not None  # Ensure container was created successfully
+            if container is None:
+                raise AssertionError("Docker container was not created successfully")
            logs = container.logs(stream=True)
            print(Rule("[bold green]Docker Logs Begin[/bold green]", style="dark_orange"))
            table = Table(title="Run Info", show_header=False)
@@ -270,6 +270,11 @@ class LoopBase:
                        msg = "We have reset the loop instance, stop all the routines and resume."
                        raise self.LoopResumeError(msg) from e
                    else:
+                        # Do NOT advance step_idx for unhandled exceptions (e.g. LoopResumeError
+                        # propagating from _propose). Keeping step_idx at the current step lets
+                        # kickoff_loop retry step 0 on the next resume instead of permanently
+                        # corrupting the loop with a missing direct_exp_gen result.
+                        step_forward = False
                        raise  # re-raise unhandled exceptions
                finally:
                    # No matter the execution succeed or not, we have to finish the following steps
@@ -30,7 +30,8 @@ def wait_retry(
    >>> counter
    2
    """
-    assert retry_n > 0, "retry_n should be greater than 0"
+    if retry_n <= 0:
+        raise ValueError("retry_n should be greater than 0")

    def decorator(f: Callable[..., ASpecificRet]) -> Callable[..., ASpecificRet]:
        def wrapper(*args: Any, **kwargs: Any) -> ASpecificRet:
@@ -84,7 +84,8 @@ class WorkflowTracker:
            # Log timer status if timer is started
            if self.loop_base.timer.started:
                remain_time = self.loop_base.timer.remain_time()
-                assert remain_time is not None
+                if remain_time is None:
+                    raise AssertionError("remain_time should not be None")
                mlflow.log_metric("remain_time", remain_time.total_seconds())
                mlflow.log_metric(
                    "remain_percent",
@@ -9,8 +9,8 @@ psutil
 fire
 fuzzywuzzy
 openai
-litellm>=1.73  # to support `from litellm import get_valid_models`
-aiohttp>=3.13.4  # CVE-2026-22815, CVE-2026-34515, CVE-2026-34516, CVE-2026-34525
+litellm>=1.83.14  # to support `from litellm import get_valid_models`
+aiohttp>=3.13.4  # CVE-2026-22815, CVE-2026-34515, CVE-2026-34516, CVE-2026-34525; >=3.13.4 due to litellm==1.83.14 exact pin
 azure.identity
 pyarrow
 rich
@@ -98,8 +98,8 @@ optuna>=3.5.0
 beautifulsoup4>=4.12.0

 # ML Training Pipeline
-lightgbm>=3.3.0
-scipy>=1.9.0
+lightgbm>=3.3.5
+scipy>=1.15.3

 # RL Trading (optional - system works without these)
 # Install for full RL training: pip install stable-baselines3[extra] gymnasium
@@ -8,7 +8,7 @@
 # Only install if you want to use full PPO/A2C/SAC training.

 # Core RL library
-stable-baselines3[extra]>=2.0.0
+stable-baselines3[extra]>=2.8.0

 # Gymnasium environment (OpenAI Gym successor)
 gymnasium>=0.29.0
@@ -198,6 +198,55 @@ def scan_factors(workspace_dir: Path, skip_evaluated: bool = True) -> List[Facto
    return factors


+# ---------------------------------------------------------------------------
+# Look-ahead bias detection for daily-constant factors
+# ---------------------------------------------------------------------------
+def _shift_daily_constant_factor_if_needed(factor_col: "pd.Series", factor_name: str) -> "pd.Series":
+    """Detect daily-constant factors (look-ahead bias) and shift by 1 trading day."""
+    sample_days = factor_col.index.get_level_values("datetime").normalize().unique()
+    if len(sample_days) < 10:
+        return factor_col
+    rng = np.random.default_rng(42)
+    days_to_check = rng.choice(sample_days, size=min(50, len(sample_days)), replace=False)
+    constant_count = 0
+    for day in days_to_check:
+        day_mask = factor_col.index.get_level_values("datetime").normalize() == day
+        day_vals = factor_col[day_mask].dropna()
+        if len(day_vals) == 0:
+            continue
+        if day_vals.nunique() == 1:
+            constant_count += 1
+    fraction_constant = constant_count / len(days_to_check)
+    if fraction_constant < 0.90:
+        return factor_col
+    # Shift by 1 trading day per instrument
+    import logging
+    logging.getLogger(__name__).info(
+        "Factor '%s' is %.0f%% daily-constant — shifting 1 trading day to fix look-ahead bias",
+        factor_name, fraction_constant * 100,
+    )
+    instruments = factor_col.index.get_level_values("instrument").unique() if "instrument" in factor_col.index.names else [None]
+    shifted_parts = []
+    for instr in instruments:
+        if instr is not None:
+            mask = factor_col.index.get_level_values("instrument") == instr
+            col_instr = factor_col[mask]
+        else:
+            col_instr = factor_col
+        dates = col_instr.index.get_level_values("datetime").normalize()
+        trading_days = dates.unique().sort_values()
+        day_first = col_instr.groupby(dates).first()
+        day_first_shifted = day_first.shift(1)
+        day_first_shifted.index = pd.to_datetime(day_first_shifted.index)
+        day_map = day_first_shifted.reindex(pd.to_datetime(trading_days)).values
+        new_vals = pd.Series(
+            day_map[np.searchsorted(trading_days.values, dates.values)],
+            index=col_instr.index,
+        )
+        shifted_parts.append(new_vals)
+    return pd.concat(shifted_parts).sort_index()
+
+
 # ---------------------------------------------------------------------------
 # Factor evaluator
 # ---------------------------------------------------------------------------
@@ -263,6 +312,7 @@ def evaluate_factor_full(factor: FactorInfo, full_data: pd.DataFrame,
            result = pd.read_hdf(str(result_file), key="data")
            total_count = len(result)
            factor_val = result.iloc[:, 0]
+            factor_val = _shift_daily_constant_factor_if_needed(factor_val, factor.factor_name)
            non_null_count = factor_val.notna().sum()

            if non_null_count < 1000:
@@ -250,7 +250,7 @@ Hard requirements:
 - NO global mean/std — always use rolling(window).mean() with shift(1) to avoid look-ahead bias"""

        else:
-            system_prompt = f"""You are a quantitative trading expert specializing in EUR/USD intraday strategies.
+            system_prompt = f"""You are a quantitative trading expert specializing in EUR/USD daily swing strategies.

 CRITICAL RULES for {STYLE_DESC} (forward horizon: {FORWARD_BARS} bars = ~{FORWARD_BARS/60:.1f} hours):
 1. ONLY use the factors listed below - no others!
@@ -258,15 +258,27 @@ CRITICAL RULES for {STYLE_DESC} (forward horizon: {FORWARD_BARS} bars = ~{FORWAR
 3. Create a pandas Series called 'signal' with values: 1 (long), -1 (short), 0 (neutral)
 4. signal.index MUST match close.index
 5. signal.name must be 'signal'
+6. IMPORTANT: factors are DAILY values broadcast to every 1-minute bar — they change once per day.
+   Use daily-level logic: compare today's factor value to a rolling daily mean (window 5-20 DAYS).
+   To get daily rolling mean: group by date, take first value per day, compute rolling, then reindex back.
+   Example: dates = factors[col].index.get_level_values('datetime').normalize()
+            daily_vals = factors[col].groupby(dates).first()
+            daily_mean = daily_vals.rolling(10).mean().shift(1)
+            daily_signal = (daily_vals > daily_mean).astype(int) * 2 - 1
+            signal = daily_signal.reindex(dates).values  (broadcast back to minute bars)
+7. The signal should change roughly once per day — this produces ~250-500 trades over 6 years.
+8. Keep conditions SIMPLE: one factor above/below its N-day rolling average. Avoid combining 3+ conditions.

 Output ONLY valid JSON with these fields:
 {{"strategy_name": "short_name", "factor_names": ["f1", "f2"], "description": "one sentence", "code": "python code"}}"""

-            user_prompt = f"""Create a EUR/USD trading strategy using these factors:
+            user_prompt = f"""Create a EUR/USD SWING trading strategy (hold ~{FORWARD_BARS/60:.0f} hours) using these factors:

 {factor_list}

-{f'Previous feedback: {feedback}' if feedback else 'First attempt - be creative!'}"""
+{f'Previous feedback: {feedback}' if feedback else 'First attempt - be creative!'}
+
+Use daily-level signal logic (factor above/below rolling daily mean). Signal changes once per day."""
        
        api = APIBackend()
        response = api.build_messages_and_create_chat_completion(
@@ -371,8 +383,8 @@ signal.fillna(0).to_pickle('signal.pkl')
        txn_cost_bps=TXN_COST_BPS,
        forward_returns=fwd_returns,
        oos_start=OOS_START_DEFAULT,
-        wf_rolling=True,
-        mc_n_permutations=200,
+        wf_rolling=False,   # too slow on 2M bars — run via rebacktest script instead
+        mc_n_permutations=50,
    )

 # ============================================================================
@@ -579,7 +591,7 @@ def main(target_count=10):
                # Check acceptance criteria — OOS must be profitable + statistically significant
                mc_ok = mc_pvalue is None or mc_pvalue < 0.20  # lenient: top 20% non-random
                wf_ok = wf_consistency is None or wf_consistency >= 0.5  # ≥50% of WF windows profitable
-                if (abs(ic) > MIN_IC and sharpe > MIN_SHARPE and trades > MIN_TRADES and dd > MAX_DRAWDOWN
+                if (abs(ic or 0) > MIN_IC and sharpe > MIN_SHARPE and trades > MIN_TRADES and dd > MAX_DRAWDOWN
                        and oos_sharpe > 0.0 and oos_monthly > 0.0 and mc_ok and wf_ok):
                    # ACCEPT
                    strategy['real_backtest'] = bt_result
@@ -636,9 +648,10 @@ def main(target_count=10):
                    oos_info = f"OOS_Sharpe={oos_sharpe:+.2f} OOS_Mon={oos_monthly:+.2f}%" if oos_sharpe is not None else ""
                    mc_info  = f" MC_p={mc_pvalue:.2f}" if mc_pvalue is not None else ""
                    wf_info  = f" WF_consistency={wf_consistency:.0%}" if wf_consistency is not None else ""
-                    _log.info(f"REJECTED  IC={ic:.4f}  Sharpe={sharpe:.2f}  Trades={trades}  DD={dd:.1%}  {oos_info}{mc_info}{wf_info}")
+                    _ic = ic or 0; _sh = sharpe or 0; _dd = dd or 0
+                    _log.info(f"REJECTED  IC={_ic:.4f}  Sharpe={_sh:.2f}  Trades={trades}  DD={_dd:.1%}  {oos_info}{mc_info}{wf_info}")
                    feedback_history.append(
-                        f"Failed: IC={ic:.4f}, Sharpe={sharpe:.2f}, Trades={trades}, DD={dd:.1%}, "
+                        f"Failed: IC={_ic:.4f}, Sharpe={_sh:.2f}, Trades={trades}, DD={_dd:.1%}, "
                        f"OOS_Sharpe={oos_sharpe:+.2f}, OOS_Monthly={oos_monthly:+.2f}%"
                        + (f", MC_p={mc_pvalue:.2f}" if mc_pvalue is not None else "")
                        + (f", WF_consistency={wf_consistency:.0%}" if wf_consistency is not None else "")
@@ -0,0 +1,249 @@
+"""Tests for FactorAutoFixer — the pre-execution code patcher."""
+
+import pytest
+
+from rdagent.components.coder.factor_coder.auto_fixer import FactorAutoFixer
+
+
+@pytest.fixture()
+def fixer():
+    return FactorAutoFixer()
+
+
+class TestResetIndexGroupby:
+    def test_replaces_level_groupby_on_reset_var(self, fixer):
+        code = "df_r = df.reset_index()\ndf_r['x'] = df_r.groupby(level=1)['$close'].mean()"
+        result = fixer.fix(code)
+        assert "groupby('instrument')" in result
+
+    def test_does_not_touch_normal_multiindex_groupby(self, fixer):
+        code = "df['x'] = df.groupby(level=1)['$close'].mean()"
+        result = fixer.fix(code)
+        assert "groupby(level=1)" in result
+
+
+class TestGroupbyMixedLevels:
+    def test_strips_string_from_mixed_list(self, fixer):
+        result = fixer.fix("df.groupby(level=[1, 'date']).apply(fn)")
+        assert "groupby(level=1)" in result
+
+    def test_multiple_ints_kept(self, fixer):
+        result = fixer.fix("df.groupby(level=[0, 1, 'x']).apply(fn)")
+        assert "groupby(level=[0, 1])" in result
+
+
+class TestGroupbyColumnOnMultiindex:
+    def test_instrument_date_becomes_two_level(self, fixer):
+        code = "df['v'] = df.groupby(['instrument', 'date'])['$volume'].cumsum()"
+        result = fixer.fix(code)
+        assert "get_level_values(1)" in result
+        assert "normalize()" in result
+        assert "level=1)" not in result.split("get_level_values")[0]
+
+    def test_date_instrument_becomes_two_level(self, fixer):
+        code = "df['v'] = df.groupby(['date', 'instrument'])['$volume'].cumsum()"
+        result = fixer.fix(code)
+        assert "get_level_values(0).normalize()" in result
+        assert "get_level_values(1)" in result
+
+    def test_single_instrument_becomes_level1(self, fixer):
+        result = fixer.fix("df.groupby(['instrument'])['x'].mean()")
+        assert "groupby(level=1)" in result
+
+    def test_reset_index_not_double_fixed(self, fixer):
+        # After reset_index fix emits groupby('instrument'), this fixer must NOT
+        # convert that to groupby(level=1).
+        code = "df_r = df.reset_index()\ndf_r['x'] = df_r.groupby(level=1)['p'].mean()"
+        result = fixer.fix(code)
+        assert "groupby('instrument')" in result
+
+
+class TestChainedGroupby:
+    def test_chained_groupby_level_then_date(self, fixer):
+        code = "df.groupby(level=1).groupby('date')['price_volume'].transform('cumsum')"
+        result = fixer.fix(code)
+        assert "get_level_values(1)" in result
+        assert "get_level_values(0).normalize()" in result
+        assert ".groupby('date')" not in result
+
+    def test_chained_groupby_with_double_quotes(self, fixer):
+        code = 'df.groupby(level=0).groupby("date")["col"].sum()'
+        result = fixer.fix(code)
+        assert "get_level_values" in result
+        assert '.groupby("date")' not in result
+
+    def test_list_with_level_keyword_syntax_error(self, fixer):
+        # groupby([level=1, 'date']) is a SyntaxError — must be fixed before execution
+        code = "asian_vol = df[mask].groupby([level=1, 'date'])['log_return'].std()"
+        result = fixer.fix(code)
+        assert "get_level_values(1)" in result
+        assert "normalize()" in result
+        assert "level=1," not in result
+
+    def test_list_with_level_keyword_reversed(self, fixer):
+        code = "df.groupby(['date', level=1])['x'].mean()"
+        result = fixer.fix(code)
+        assert "get_level_values" in result
+        assert "level=1" not in result
+
+
+class TestMinPeriodsNotTouched:
+    def test_small_min_periods_preserved(self, fixer):
+        # _fix_min_periods is disabled — LLM-set min_periods must not be changed.
+        # window=60, min_periods=1 should stay as-is (was wrongly raised to 60 before).
+        result = fixer.fix("df.groupby(level=1)['x'].transform(lambda x: x.rolling(window=60, min_periods=1).mean())")
+        assert "min_periods=1" in result
+
+    def test_large_window_min_periods_preserved(self, fixer):
+        # window=240 > 96 bars/day: if min_periods were set to 240 the output would be
+        # all-NaN for intraday data. Verify we leave it untouched.
+        result = fixer.fix("df['x'] = df.groupby(level=1)['y'].transform(lambda x: x.rolling(240, min_periods=10).std())")
+        assert "min_periods=10" in result
+
+
+class TestInstrumentColumnAccess:
+    def test_instrument_column_replaced(self, fixer):
+        code = "df['group_key'] = df['instrument'] + '_' + df['day_id'].astype(str)"
+        result = fixer.fix(code)
+        assert "df.index.get_level_values(1)" in result
+        assert "df['instrument']" not in result
+
+    def test_reset_index_var_not_touched(self, fixer):
+        # After reset_index, 'instrument' IS a real column — must not be replaced
+        code = "df_r = df.reset_index()\nval = df_r['instrument'].unique()"
+        result = fixer.fix(code)
+        assert "df_r['instrument']" in result
+        assert "get_level_values" not in result
+
+    def test_groupby_after_instrument_fix(self, fixer):
+        # Combined: df['instrument'] in a groupby context
+        code = "df['key'] = df['instrument']\nout = df.groupby(df['key'])[['$close']].mean()"
+        result = fixer.fix(code)
+        assert "df['instrument']" not in result
+
+    def test_assignment_target_not_touched(self, fixer):
+        # df['instrument'] = <expr> is an assignment — must NOT be converted to
+        # df.index.get_level_values(1) = <expr> (SyntaxError)
+        code = "df['instrument'] = df.index.get_level_values('instrument')"
+        result = fixer.fix(code)
+        assert "df['instrument'] =" in result
+
+
+class TestInstrumentLocMultiindex:
+    def test_loc_replaced_with_xs(self, fixer):
+        code = (
+            "for instrument in df.index.get_level_values('instrument').unique():\n"
+            "    inst_df = df.loc[instrument].copy()\n"
+        )
+        result = fixer.fix(code)
+        assert "df.xs(instrument, level=1)" in result
+        assert "df.loc[instrument]" not in result
+
+    def test_loc_replaced_with_level1_int(self, fixer):
+        code = (
+            "for inst in df.index.get_level_values(1).unique():\n"
+            "    data = df.loc[inst]\n"
+        )
+        result = fixer.fix(code)
+        assert "df.xs(inst, level=1)" in result
+
+    def test_loc_assignment_not_touched(self, fixer):
+        # Write-back df.loc[instrument] = ... must not be changed
+        code = (
+            "for instrument in df.index.get_level_values('instrument').unique():\n"
+            "    df.loc[instrument] = modified\n"
+        )
+        result = fixer.fix(code)
+        assert "df.loc[instrument] = modified" in result
+
+    def test_non_instrument_loop_not_touched(self, fixer):
+        # for-loop not related to instrument levels must not be changed
+        code = "for date in dates:\n    sub = df.loc[date]\n"
+        result = fixer.fix(code)
+        assert "df.loc[date]" in result
+
+
+class TestGroupbyLevelStringNames:
+    def test_level_instrument_date_replaced(self, fixer):
+        code = "df.groupby(level=['instrument', 'date'])['col'].transform('sum')"
+        result = fixer.fix(code)
+        assert "get_level_values(1)" in result
+        assert "get_level_values(0).normalize()" in result
+        assert "level=['instrument', 'date']" not in result
+
+    def test_level_date_instrument_replaced(self, fixer):
+        code = "data.groupby(level=['date', 'instrument'])['x'].mean()"
+        result = fixer.fix(code)
+        assert "get_level_values(0).normalize()" in result
+        assert "get_level_values(1)" in result
+
+    def test_level_instrument_single_replaced(self, fixer):
+        code = "df.groupby(level=['instrument'])['vol'].sum()"
+        result = fixer.fix(code)
+        assert "groupby(level=1)" in result
+        assert "level=['instrument']" not in result
+
+
+class TestGroupbyApplyToTransform:
+    def test_col_apply_lambda_replaced(self, fixer):
+        code = "df_overlap.groupby(level=1)['$close'].apply(lambda x: np.log(x / x.shift(1)))"
+        result = fixer.fix(code)
+        assert ".transform(" in result
+        assert ".apply(" not in result
+
+    def test_col_apply_lambda_preserves_lambda_body(self, fixer):
+        code = "series.groupby(level=1)['ret'].apply(lambda x: x.cumsum())"
+        result = fixer.fix(code)
+        assert "lambda x: x.cumsum()" in result
+        assert ".transform(" in result
+
+    def test_transform_reset_index_stripped(self, fixer):
+        # .transform() already preserves index — .reset_index() after it is wrong
+        code = "df['v'] = df.groupby(level=1)['x'].transform(lambda x: x.rolling(20).mean()).reset_index(level=0, drop=True)"
+        result = fixer.fix(code)
+        assert ".reset_index(level=0, drop=True)" not in result
+        assert ".transform(" in result
+
+
+class TestZeroVolumeProxy:
+    def test_injects_proxy_when_volume_used(self, fixer):
+        code = (
+            "def calc():\n"
+            "    df = pd.read_hdf('data.h5', key='data')\n"
+            "    df['pv'] = df['$close'] * df['$volume']\n"
+            "    return df[['pv']]\n"
+        )
+        result = fixer.fix(code)
+        assert "volume proxy" in result
+        assert "df['$volume'] = df['$high'] - df['$low']" in result
+        # Proxy must come right after read_hdf line
+        lines = result.splitlines()
+        hdf_idx = next(i for i, l in enumerate(lines) if "read_hdf" in l)
+        assert "volume proxy" in lines[hdf_idx + 1]
+
+    def test_no_injection_when_volume_absent(self, fixer):
+        code = "df = pd.read_hdf('data.h5', key='data')\ndf['x'] = df['$close'].pct_change()\n"
+        result = fixer.fix(code)
+        assert "volume proxy" not in result
+
+    def test_no_double_injection(self, fixer):
+        code = (
+            "def calc():\n"
+            "    df = pd.read_hdf('data.h5', key='data')\n"
+            "    # volume proxy: $volume is always 0 in FX data — use price-range as proxy\n"
+            "    if (df['$volume'] == 0).all():\n"
+            "        df['$volume'] = df['$high'] - df['$low']\n"
+            "    df['pv'] = df['$close'] * df['$volume']\n"
+        )
+        result = fixer.fix(code)
+        assert result.count("volume proxy") == 1
+
+
+class TestRollingDdof:
+    def test_removes_ddof_from_rolling_args(self, fixer):
+        result = fixer.fix("df.rolling(20, min_periods=1, ddof=1).std()")
+        assert "ddof" not in result
+
+    def test_removes_ddof_from_std_args(self, fixer):
+        result = fixer.fix("df.rolling(20).std(ddof=1)")
+        assert "ddof" not in result