The Agent Skills Directory

Privilege Escalation (HIGH): The file SKILL.md includes instructions for executing sudo commands (sudo apt-get install, sudo systemctl start) which allows an agent to perform system-level administrative operations without restricted oversight.
Indirect Prompt Injection (HIGH):
Ingestion points: The skill reads data directly from MongoDB and PostgreSQL databases via the psql and mongosh CLI tools.
Boundary markers: There are no markers or specific instructions to isolate or ignore malicious command strings embedded within database records.
Capability inventory: The skill allows for direct modification of database schemas, user permissions, and data records via aggregation pipelines and SQL queries.
Sanitization: No evidence of input validation or parameterization of database queries is present, allowing data to be treated as executable code.
Metadata Poisoning (MEDIUM): The scripts/requirements.txt file incorrectly states that the skill uses only the standard library, while test_db_migrate.py confirms that the skill depends on pymongo and psycopg2 for its core functionality.
External Downloads (MEDIUM): SKILL.md and scripts/requirements.txt recommend downloading binary tools from external sources (mongodb.com, postgresql.org) without providing integrity hashes or verification steps.

databases