The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill grants the agent the ability to terminate arbitrary system processes by PID (./scripts/god/powers.sh terminate pid:12345). This can be used to disable security tools or disrupt critical system services.
[COMMAND_EXECUTION] (HIGH): The skill provides a modify command that allows the agent to append, prepend, or replace the contents of any file (./scripts/god/powers.sh modify "path/to/file" replace "new content"). This capability allows an agent to inject malicious code into existing scripts, configuration files, or shell profiles (e.g., .bashrc), leading to persistent system compromise.
[COMMAND_EXECUTION] (HIGH): Provides capabilities for repository-wide destructive actions, including hard git resets (./scripts/god/powers.sh rollback HEAD~3 hard) and system-wide shutdowns (./scripts/god/powers.sh terminate all).
[PROMPT_INJECTION] (MEDIUM): The 'God Committee' and 'God Intervention' framing acts as a high-privilege role-play scenario. This may lead the agent to believe it has 'overriding' authority, potentially bypassing safety guardrails when processing user requests that use the intervention triggers.
[DATA_EXFILTRATION] (MEDIUM): The skill enables the agent to read and modify sensitive project configurations (e.g., .god/config.json). While no direct external network calls are listed, the unrestricted file system access provided by the repair and modification tools allows for sensitive data discovery.

god-intervention