The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingest and processes external JSON schema files through several bundled Python scripts, which creates a vulnerability surface where malicious instructions could be embedded in schema metadata or description fields.
Ingestion points: The --schema parameter in scripts/generate_sample_data.py, scripts/migrate_schema.py, and scripts/validate_schema.py allows the skill to load user-supplied JSON files.
Boundary markers: The skill instructions do not provide specific delimiters or ignore-instructions markers to help the agent distinguish between data and potentially malicious natural language instructions within the schema.
Capability inventory: The skill environment is configured with allowed-tools that include file system Write access and Bash execution for various database shells (mongosh, mysql, psql).
Sanitization: The scripts use standard JSON loading libraries but do not include explicit logic to sanitize or filter for prompt injection patterns within the processed content.
[SAFE]: The skill exclusively references documentation from well-known and established technology providers.
Evidence: Resource links in SKILL.md point to official documentation sites for MongoDB, Amazon Web Services (DynamoDB), and Redis, which are trusted services.

modeling-nosql-data