tutor
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions (e.g., in
templates/AGENTS.lab1.md) direct the AI agent to automatically download and install the skill from the author's GitHub repository (https://github.com/DyingCoderLin/OS-Lab-Tutor) usinggit cloneif the skill is not found locally. This is categorized as a safe vendor-resource download. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted student-provided data, such as C source code and lab reports, which exposes the agent to indirect prompt injection. Malicious instructions placed in these files could potentially manipulate the agent's behavior. Ingestion points: Reads student-written code files (e.g.,
buddy.c,uart.c) and reports at student-designated paths. Boundary markers: Lacks explicit delimiters or 'ignore' instructions when reading external file content into the prompt context. Capability inventory: Possesses capabilities to execute shell commands (make,tar,git) and read file system content. Sanitization: No sanitization or validation of the ingested file content is performed. - [COMMAND_EXECUTION]: The agent is authorized to run build and test commands (specifically
make,make grade, andtar) within the user's project directory. This capability allows the agent to execute code defined in the student'sMakefile, which could be exploited if theMakefilecontains malicious logic.
Audit Metadata