vectorcraft-cli-paths
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes Python scripts (
bootstrap.py,cli.py,commands.py,workflow.py,desktop.py) from thevectorcraft-harnessskill to drive vector operations and handle setup tasks. - [EXTERNAL_DOWNLOADS]: The skill downloads and installs a native CLI and desktop application from vendor-controlled URLs during the first-use initialization. These downloads are verified using fixed checksums (SHA-256) as described in the documentation.
- [DYNAMIC_EXECUTION]: Vector editing tasks are defined in JSON-based plans (
craft-command-plan/v1), which are parsed and executed by the harness and native CLI at runtime. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted external assets (SVG, PNG, JPG, PDF) using the
asset.placeanddocument.opencommands. While the skill documentation mentions sanitization steps—such as filtering SVG scripts and foreignObject elements—the processing of these files by the CLI could allow malicious metadata or structures to enter the agent's context. The capability to execute subprocesses and write files further increases this surface, although it is mitigated by the described filesystem sandboxing policies.
Audit Metadata