~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
Python academiaanthropicarxivbravedeep-researchencryptionhome-automationhomeserverlocallocal-deep-researchlocal-llmmistralollamaopenaipubmedresearchresearch-toolretrieval-augmented-generationsearxngself-hosted
fix(security): suppress Bearer SQL-injection false positive in migration 0018 (alert #7993) (#4788)
LearningCircuita7214633df0f
diff --git a/src/local_deep_research/database/migrations/versions/0018_remove_mcp_strategy.py b/src/local_deep_research/database/migrations/versions/0018_remove_mcp_strategy.pyindex 35c1e536d1..f1c4dc78db 100644--- a/src/local_deep_research/database/migrations/versions/0018_remove_mcp_strategy.py+++ b/src/local_deep_research/database/migrations/versions/0018_remove_mcp_strategy.py@@ -72,7 +72,11 @@ _queued = sa.table(# The two benchmark tables share the same shape. They are declared here with# literal names (rather than building ``sa.table(table_name, ...)`` from a# passed-in string) so the migration never constructs a SQL identifier from a-# variable — there is no dynamic-table-name flow for a static analyzer to flag.+# variable. The shared helper below still receives the table object as a+# parameter, which Bearer's taint analysis reports as a SQL-injection false+# positive on its ``conn.execute`` call — suppressed with a+# ``# bearer:disable python_lang_sql_injection`` directive placed on the line+# directly above the statement (the only placement Bearer honors)._benchmark_runs = sa.table("benchmark_runs", sa.column("id"), sa.column("search_config"))@@ -184,9 +188,20 @@ def _rewrite_benchmark_search_configs(conn, table) -> None:"""Rewrite removed strategies in a benchmark table's ``search_config`` JSON.``table`` is one of the module-level literal ``sa.table`` objects- (``_benchmark_runs`` / ``_benchmark_configs``) — never a string — so the- query below contains no dynamically-built SQL identifier.+ (``_benchmark_runs`` / ``_benchmark_configs``) — never a string and never+ external input — so the table/column identifiers are fixed literals that+ SQLAlchemy renders verbatim (and the UPDATE binds its value as a parameter);+ nothing user-derived reaches the SQL. Bearer's taint analysis nonetheless+ reports the ``table`` parameter flowing into ``conn.execute`` as SQL+ injection (a false positive). It is suppressed with a+ ``# bearer:disable python_lang_sql_injection`` directive on the line+ directly above the statement — the only placement Bearer honors (a+ same-line trailing directive is ignored). Bearer flags only the read; the+ write below carries the same directive defensively."""+ # ``table`` is a module-level literal sa.table (see docstring) — the Bearer+ # SQL-injection report on this read is a false positive.+ # bearer:disable python_lang_sql_injectionrows = conn.execute(sa.select(table.c.id, table.c.search_config)).fetchall()rewritten = 0@@ -198,6 +213,8 @@ def _rewrite_benchmark_search_configs(conn, table) -> None:continueif config.get("search_strategy") in REMOVED_STRATEGIES:config["search_strategy"] = REPLACEMENT_STRATEGY+ # Same module-level literal table object, no external input.+ # bearer:disable python_lang_sql_injectionconn.execute(sa.update(table).where(table.c.id == row_id)