The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/result_ranker.py uses subprocess.run() with shell=True to execute evaluation commands defined in the session configuration. This allows for arbitrary shell command execution within the agent's worktree environment.- [DYNAMIC_EXECUTION]: The skill facilitates the execution of arbitrary commands provided via the --eval-cmd parameter. These commands are stored in the session configuration and executed during the ranking phase, creating a potential for code execution if the configuration is controlled by an untrusted source.- [INDIRECT_PROMPT_INJECTION]: The skill lacks sanitization for user-provided task descriptions and evaluation commands, making it vulnerable to indirect injection.
Ingestion points: Task descriptions and evaluation parameters are directly interpolated into the sub-agent dispatch prompts in SKILL.md and references/agent-templates.md.
Boundary markers: There are no delimiters or instructions to prevent the agent from obeying instructions embedded in the task data.
Capability inventory: The skill and its sub-agents can execute shell commands, perform git operations, and modify the file system.
Sanitization: No input validation is performed on task descriptions or commands before prompt interpolation or execution.

agenthub