The Agent Skills Directory

Dynamic Code Generation and Execution: The skill authored by the vendor is designed to generate and run reproduction scripts (e.g., Python, C, shell commands) to verify security vulnerabilities. This represents the core functionality of the tool and is supported by explicit instructions to use compilers, debuggers, and sandbox environments.
Environment Isolation Constraints: To mitigate the risks of executing generated payloads, the skill mandates a "Strict Host Isolation Constraint." It forbids running commands directly on the host terminal and requires all executions to occur within isolated containers or sandboxes (e.g., Docker, QEMU).
Indirect Prompt Injection Surface: The skill processes external, potentially untrusted content from the target codebase to identify and verify trigger paths. This ingestion of external data constitutes a surface for indirect prompt injection, where malicious content within the analyzed files could attempt to influence the agent's behavior. This risk is addressed by the skill's restricted execution environment.
Automated Workspace Management: The skill performs programmatic updates to JSON findings and state files. It uses atomic file operations and locking mechanisms to ensure data integrity during these updates, preventing race conditions or file corruption.

mantis-reproduce