lookml-field-curation
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from database schemas and LookML files to generate refinements, creating a surface for indirect prompt injection.
- Ingestion points: Metadata from
INFORMATION_SCHEMA(column names, types, and comments) and local LookML field definitions. - Boundary markers: Output refinements are formatted within LookML view refinement blocks (
view: +name { ... }). - Capability inventory: Subprocess execution of Python scripts via
uv runand database queries via the platform'slkr code-mode sandboxenvironment. - Sanitization: The scripts
profile_string_fields.pyandenrich_partition_clustering.pyexplicitly sanitize generated description strings by escaping backslashes, double quotes, and newline characters to prevent malformed LookML generation. - [SAFE]: All script executions and tool invocations rely on local files and platform-authorized CLI tools (
lkr). The skill does not exhibit any patterns of data exfiltration, unauthorized credential access, or obfuscation.
Audit Metadata