The Agent Skills Directory

[COMMAND_EXECUTION]: The script validate-sql-syntax.sh invokes psql -f on SQL files to validate their syntax. The psql utility parses meta-commands like \\!, which allows for shell command execution. This means a malicious SQL file can execute commands in the agent's environment if the file is processed during validation.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes SQL migration files and reflects names and code snippets into a markdown validation report. Because the agent reads this report to provide feedback, malicious instructions embedded in the SQL content could lead to indirect prompt injection.
Ingestion points: SQL files analyzed by full-validation.sh and its associated scripts in the scripts/ directory.
Boundary markers: The validation report lacks delimiters or specific instructions to help the agent distinguish between data and instructions when reading the results.
Capability inventory: Subprocess execution (Bash), file write operations, and psql command execution.
Sanitization: The skill does not sanitize or escape SQL identifiers or comments before placing them in the final markdown report.
[DYNAMIC_EXECUTION]: The syntax validation process relies on psql, an interpreter that executes the instructions and meta-commands within the SQL files. This dynamic interpretation of code from untrusted project files is a significant attack vector.

schema-validation