The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains strong instructional directives like 'MANDATORY CHECKPOINT' and 'STOP,' which are designed to enforce human oversight and prevent the agent from spawning sub-agents autonomously without explicit user consent. These act as safety guardrails rather than malicious overrides.- [COMMAND_EXECUTION]: The skill utilizes standard version control and text processing utilities, such as git and grep, to manage development branches and analyze file dependencies. These commands are applied correctly within the scope of parallel task management and integration.- [DATA_EXFILTRATION]: Local directory usage (e.g., /tmp/) for 'disk-mediated dispatch' allows agents to share context without external network operations. No patterns for exfiltrating sensitive data to remote servers were identified.

parallel