sendcutsend
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches current manufacturing guides and engineering specifications directly from
cdn.sendcutsend.com. These references are used as the ground truth for manufacturing readiness reports. - [COMMAND_EXECUTION]: Executes Python code using the
build123dandezdxflibraries to perform local geometry inspections, such as measuring part dimensions, hole sizes, and flange depths from user-provided CAD files. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from both user-uploaded CAD files (DXF/STEP) and external web sources (JSON/Markdown from SendCutSend).
- Ingestion points: CAD files (DXF/STEP) in
SKILL.mdand remote configuration files defined inreferences/official-sources.md. - Boundary markers: Instructions specify that source data should be treated as "evidence feeds" rather than executable logic, and the agent is cautioned to handle missing or conflicting data without assuming success.
- Capability inventory: The skill uses Python for geometric analysis and generates reports based on comparisons.
- Sanitization: The skill mandates defensive parsing (e.g., handling numeric strings, size strings, and
N/Avalues) to ensure data integrity during comparison.
Audit Metadata