The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides multiple reference implementations that ingest untrusted user data from query parameters to dynamically construct database queries. For example, references/advanced-filter-parser.md and references/python-filtering-sqlalchemy.md demonstrate patterns for mapping user-supplied keys and values to database operators like MongoDB's $regex or SQLAlchemy's attribute access via getattr. This creates an attack surface where an attacker could potentially perform Regular Expression Denial of Service (ReDoS) or access unauthorized fields if the developer does not implement the whitelisting strategies recommended in the skill's Best Practices section.\n
Ingestion points: Query parameters are ingested in references/advanced-filter-parser.md (via URLSearchParams), references/python-filtering-sqlalchemy.md (via request.args), and references/query-parameter-filtering.md (via req.query).\n
Boundary markers: No explicit delimiters or instructions are used within the parsing logic to separate user data from query logic, increasing the risk of injection.\n
Capability inventory: The skill enables the agent to generate code that performs database read operations via MongoDB (find, countDocuments), SQLAlchemy (filter, order_by), and Elasticsearch (search).\n
Sanitization: While references/query-parameter-filtering.md includes an allowedFilters whitelist, the more generic implementations in references/advanced-filter-parser.md and references/python-filtering-sqlalchemy.md lack built-in validation, relying entirely on external enforcement.

api-filtering-sorting