The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands to manage its workspace and ensure task isolation. It instructs the agent to create temporary directories using mktemp and manage parallel code candidates using git worktree and git branch commands.
[INDIRECT_PROMPT_INJECTION]: The workflow distributes a single task prompt to multiple sub-agents (candidates), creating a surface where malicious instructions in the input could affect all sub-agents simultaneously.
Ingestion points: Phase A (common prompt for all candidates).
Boundary markers: Isolation is achieved at the filesystem level through git worktrees, but no explicit prompt-level delimiters are specified to separate user data from candidate instructions.
Capability inventory: The skill enables spawning background agents with shell execution capabilities and performing filesystem operations.
Sanitization: No specific sanitization or validation logic for the distributed prompts is provided in the skill.

arena