grant-thinking-cn-biology
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the agent to check for and download code updates from the vendor's repository (https://github.com/Agents365-ai/365-skills) using the
git pullcommand. This mechanism is governed by user consent and is a standard update procedure for the vendor's resources. - [COMMAND_EXECUTION]: The skill utilizes shell commands including
git,awk,sed,sort, andtailto verify versions and update its own instructions, which requires the underlying agent to have shell or terminal access. - [INDIRECT_PROMPT_INJECTION]: The skill analyzes user-supplied grant concepts and proposal drafts as its primary function. Because the skill instructions authorize the use of shell commands (for the update feature), it creates a vulnerability surface where malicious instructions embedded in a user's research idea could attempt to manipulate the agent's actions.
- Ingestion points: Processing of user-provided research titles, grant ideas, and draft structures within
SKILL.md. - Boundary markers: The instructions lack specific delimiters or isolation markers to protect the agent from instructions hidden within the user-supplied data.
- Capability inventory: Active shell command execution capabilities for version management and updates.
- Sanitization: No input validation or sanitization routines are specified for handling external research content.
Audit Metadata