The Agent Skills Directory

COMMAND_EXECUTION (SAFE): The script scripts/test_warrior.sh executes the pmars command to run simulations. This is a legitimate use of the tool within the CoreWars domain and does not involve arbitrary command execution or shell injection vulnerabilities.
DATA_EXFILTRATION (SAFE): There are no network-capable commands (like curl or wget) or access to sensitive system paths (like ~/.ssh or .env). The skill operates entirely on local Redcode files.
PROMPT_INJECTION (SAFE): The markdown instructions in SKILL.md are focused on strategy development and do not contain patterns aimed at overriding the agent's safety protocols or system prompts.
EXTERNAL_DOWNLOADS (SAFE): The skill does not perform any external package installations (e.g., via pip or npm) or remote script execution.
INDIRECT_PROMPT_INJECTION (LOW): The skill ingests untrusted Redcode assembly files. While these files are processed by the pmars simulator, the agent is encouraged to read and analyze them. This creates a surface for indirect prompt injection if an attacker provides a malicious warrior file containing instructions for the LLM. However, the risk is inherent to the task of code analysis and is considered low.

winning-avg-corewars