library
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute local Python scripts (
library_file_transfer.pyandlibrary_download.py) viapython3. These scripts are part of the skill's package and facilitate file operations and metadata management within the local workspace. - [INDIRECT_PROMPT_INJECTION]: The skill interacts with external data from the Library connector, creating a surface for indirect prompt injection.
- Ingestion points: Data enters the workflow through Library tools such as
list,search,read, andfindas specified in the Routing Rules. - Boundary markers: While explicit delimiters for content bodies are not defined, the skill utilizes specific routing logic to distinguish between Library-backed and local files.
- Capability inventory: The skill has the ability to write to the filesystem (
create_library_file,replace_library_file) and execute shell commands through local scripts. - Sanitization: The skill mandates high-security data handling by instructing the agent to 'Never interpolate returned fields into shell arguments' and requiring the use of stdin heredocs for passing JSON data to scripts.
Audit Metadata