The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill defines a "prototyper" role that creates isolated codebases in the system's temporary directory to test hypotheses. This involves generating and running scripts using languages such as Node.js, Python, and Rust.
[REMOTE_CODE_EXECUTION]: The prototyping methodology explicitly permits sub-agents to install software dependencies from public package registries (e.g., via npm install, pip install) and execute the resulting code, which represents a managed but inherent RCE risk.
[COMMAND_EXECUTION]: The orchestrator and its sub-agents utilize various shell commands for project management and environment inspection, including ls, mkdir, rsync, and date. The rsync command is used to copy user project files into the temporary execution environment.
[DATA_EXFILTRATION]: Sub-agents have read-only access to the local project files and are authorized to use web search and package managers. Although instructed to limit network activity to legitimate behavior, this creates a potential path for local information to be transmitted externally during research or prototype execution.
[INDIRECT_PROMPT_INJECTION]: The skill's core function is to research external sources (web content and documentation) and local project files. Maliciously crafted content in these sources could influence the behavior of sub-agents that have shell and network access, potentially leading to unintended command execution.

afk-exploration