oa-sim
Warn
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes several system-level commands to compile and run code, including
g++,gcc,javac,python3,node, andgo build. It also executes an external local script namedquestions.pyto fetch assessment data. - [DYNAMIC_EXECUTION]: The agent is instructed to dynamically create source files and then compile or interpret them at runtime. It utilizes a specialized Python runner (defined in
RUNNER.md) that leveragessubprocess.runto execute these commands with specific time limits. - [INDIRECT_PROMPT_INJECTION]: The skill's workflow involves ingesting and executing code written by the user (the "candidate"), which represents an untrusted data source entering an execution context.
- Ingestion points: User-written solution files (
qN.<ext>) located in the session-specific candidate folder. - Boundary markers: The instructions do not specify the use of containerization or sandboxing (e.g., Docker) to isolate the execution of the user's code from the host system.
- Capability inventory: The skill possesses the ability to create directories, write files, and execute arbitrary shell commands via the compilation and execution logic.
- Sanitization: There is no evidence of sanitization, static analysis, or safety validation performed on the user-provided code prior to its execution.
Audit Metadata