The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user requirements via the $ARGUMENTS placeholder to generate sensitive database scripts and automation code.
Ingestion points: The $ARGUMENTS variable in SKILL.md and resources/implementation-playbook.md serves as the entry point for user-defined migration needs.
Boundary markers: No boundary markers or instructions are present to ensure the agent distinguishes between its instructions and potentially malicious requirements.
Capability inventory: The skill is granted Bash, Write, Edit, and Read tools, providing it the ability to create and execute the generated SQL, Python, and Shell scripts.
Sanitization: There is no evidence of sanitization or validation of the input provided through $ARGUMENTS.
[DYNAMIC_EXECUTION]: The skill provides code templates in resources/implementation-playbook.md that utilize unsafe string interpolation for SQL construction, which can lead to SQL injection vulnerabilities.
Evidence: The Python ParallelMigrator class demonstrates unsafe table name interpolation: f"INSERT INTO v2_{table_name} ..." and f"SELECT ... FROM {table_name}".
Evidence: The validate_post_migration function similarly interpolates table names directly into queries: f"SELECT COUNT(*) FROM {table['name']}".
[COMMAND_EXECUTION]: The skill includes a shell script template for database rollbacks that is susceptible to injection if parameters are not carefully controlled.
Evidence: In resources/implementation-playbook.md, the rollback_migration.sh script uses shell variables $MIGRATION_VERSION and $DATABASE directly in commands like pg_dump -d $DATABASE and psql -d $DATABASE -f ... without validation or escaping.

database-migrations-sql-migrations