vector-index-tuning
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external documentation and local repository files to inform code modifications and performance tuning.\n
- Ingestion points: Data returned by
sandbase_run, external documentation sources, and existing repository source code.\n - Boundary markers: The instructions do not define specific delimiters or instructions to ignore embedded commands within the ingested data.\n
- Capability inventory: The skill modifies codebase artifacts and executes local build tools, tests, and benchmarks.\n
- Sanitization: There is no evidence of sanitization or filtering applied to external or repository content before processing.\n- [EXTERNAL_DOWNLOADS]: The skill communicates with the SandBase API (vendor-owned) to fetch documentation and capability metadata. The workflow includes a mandatory check to show the user the endpoint, arguments, and estimated cost before performing actions.\n- [COMMAND_EXECUTION]: The skill executes local build checks, static analysis tools, and performance benchmarks as part of its verification workflow.
Audit Metadata