The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the mssql-python package, which is not a standard or recognized library for SQL Server in the Python ecosystem (unlike pyodbc or pymssql). Using unknown packages from public registries poses a risk of supply-chain attacks or dependency confusion.
[COMMAND_EXECUTION]: The scripts/reconcile.py script is vulnerable to SQL injection. It uses Python f-strings to interpolate user-provided table and column names directly into SQL queries within the extract_table and extract_hashes functions, allowing for the execution of unauthorized database commands.
[PROMPT_INJECTION]: There is a discrepancy between the skill's documented safety constraints and its actual implementation. The SKILL.md documentation instructs the agent to avoid f-string interpolation for SQL, yet the provided script relies on it for data extraction. This inconsistency can mislead an agent's assessment of the skill's safety.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface.
Ingestion points: Table and column names ingested via command-line arguments in scripts/reconcile.py and subsequently used in database queries.
Boundary markers: Absent. The script does not perform validation or sanitization on the inputs before using them in SQL construction.
Capability inventory: The skill has the capability to execute queries on two SQL Server instances using the mssql-python driver.
Sanitization: Absent for the data extraction and hashing functions, which use unsafe string interpolation despite documented constraints suggesting otherwise.

sql-server-table-reconciliation