event-talk-selection
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill consists exclusively of instructional Markdown documentation and a JSON evaluation suite. No executable scripts (Python, Node.js, Shell), system binaries, or configuration files that grant tool access were found.
- [SAFE]: No network operations, external resource downloads, or hardcoded credentials were detected in any of the analyzed files.
- [INDIRECT_PROMPT_INJECTION]: The skill functions by processing external, untrusted data (CFP submissions and talk proposals). This creates a theoretical surface for indirect prompt injection if a submission were to contain malicious instructions. However, the skill provides no active capabilities—such as file writing, shell access, or network communication—that could be leveraged by such an attack.
- Ingestion points: Talk proposals and CFP submissions ingested for review (SKILL.md, references/committee-models-and-rubrics.md).
- Boundary markers: Absent; the instructions do not specify delimiters for separating untrusted proposal content from agent instructions.
- Capability inventory: None. The skill does not include any code or invoke any tools.
- Sanitization: Absent; the instructions do not suggest filtering or sanitizing submission text.
- [SAFE]: The skill includes cross-references to other modular components within the same vendor collection (e.g., samber/dev-event-organizer-skills@event-cfp-design), which is standard practice for modular design and poses no risk.
Audit Metadata