directus-insights
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for creating Directus Insights dashboards and panels that ingest untrusted data through variables and filters. This creates a surface where malicious input could potentially influence agent behavior or dashboard rendering.
- Ingestion points: Dashboard variables (e.g.,
variable,relational-variable) and panel filter objects defined inreferences/panels.mdandreferences/building.md. - Boundary markers: No specific delimiters or safety instructions are provided to the agent to treat interpolated variable values as untrusted data.
- Capability inventory: The skill demonstrates the use of REST API calls (via
curlinreferences/building.md) to create and modify dashboards and panels, which constitutes a write capability to the connected Directus instance. - Sanitization: The documentation explicitly states that "The API saves any options unchecked," which confirms a lack of server-side validation or sanitization for panel configuration payloads.
- [COMMAND_EXECUTION]: The documentation includes shell command examples using
curlandjqto interact with the Directus REST API. While these are presented as educational examples for the user/agent to follow, they involve network operations and authentication token handling. - [EXTERNAL_DOWNLOADS]: The skill references and fetches information from external sources, including official Directus documentation domains and GitHub repositories.
- Evidence: References to
directus.comfor documentation andgithub.com/directus/*for source code and configurations. These resources are vendor-controlled and used for legitimate configuration and documentation purposes.
Audit Metadata