The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the $ARGUMENTS variable directly to construct file paths (e.g., .claude/agents/$ARGUMENTS-farmer.md) and subagent names (e.g., $ARGUMENTS-farmer). Without explicit sanitization of this input, a user could provide path traversal strings (like ../../some_file) to check for the existence of files outside the intended directory or manipulate which subagent is invoked.
[PROMPT_INJECTION]: This skill presents an indirect prompt injection surface. It retrieves and executes instructions from external files in the .claude/agents/ directory.
Ingestion points: The skill reads file names and content from the .claude/agents/ path based on the $ARGUMENTS input.
Boundary markers: No boundary markers or 'ignore embedded instructions' warnings are present when the subagent is invoked.
Capability inventory: The skill uses the agent tool to launch sub-processes with the prompt 'Execute your farming instructions now'.
Sanitization: The skill validates the existence of the file but does not sanitize the content of the farmer instruction files or the input path itself.

farm