The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the processing of external data sources for interpolation into Jinja2 templates, which represents a potential injection surface.
Ingestion points: Data is ingested from the NetBox database (ORM objects), hierarchical Config Contexts, and user-supplied POST body data via the /render-config/ and /render/ API endpoints, as described in SKILL.md and references/context-variables.md.
Boundary markers: The system utilizes standard Jinja2 delimiters (e.g., {{ ... }} and {% ... %}) to separate template logic from data.
Capability inventory: Rendered templates can perform read-only ORM queries against the NetBox database and access whitelisted system environment variables via the env filter. The skill notes that templates are restricted from file system access or arbitrary Python imports.
Sanitization: NetBox utilizes jinja2.sandbox.SandboxedEnvironment for rendering. The documentation highlights security controls such as the environment_params allowlist and the JINJA_ENVIRONMENT_PARAMS configuration for environment variable access.
[DYNAMIC_EXECUTION]: The skill provides instructions on defining and executing Jinja2 templates, which is a form of dynamic execution for configuration generation.
Evidence: SKILL.md and references/api-rendering-patterns.md detail the API methods for rendering templates and include examples of embedded logic and ORM queries.
Context: The documentation includes explicit warnings regarding performance impacts of unbounded queries and references security patches (such as CVE-2026-29514) related to template environment parameters.

netbox-config-templates