content-pipeline
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses shell commands (
ls,mv,rmdir) to organize generated files into timestamped directories and reconcile outputs from different pipeline stages. These operations are confined to the current working directory. - [COMMAND_EXECUTION]: A Python script is executed to validate the XML structure of the generated SVG cover image, ensuring it is a valid document using the
xml.etree.ElementTreelibrary. - [COMMAND_EXECUTION]: The workflow involves executing local export scripts (for PDF and MP4 generation) from a sibling skill,
carousel-creator, within the agent's environment. - [PROMPT_INJECTION]: The skill processes untrusted user input (topics and ideas) and interpolates them into content generation prompts, creating a surface for indirect prompt injection.
- Ingestion points: User-provided topic and key points collected during the intake step.
- Boundary markers: None explicitly defined in the instructions for isolating user content.
- Capability inventory: File system write access, shell command execution, and subagent invocation.
- Sanitization: The skill relies on the underlying model's safety guardrails and does not implement custom sanitization or escaping for input strings.
Audit Metadata