The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a command tool that allows the agent to execute shell commands within its assigned workspace. This is a powerful capability intended for task completion, though the instructions state it is restricted by worktree boundaries and claim validation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external sources, creating a potential surface for indirect prompt injection.
Ingestion points: The agent retrieves instructions, task context, and messages from other workers via read_context, read_messages, and pending_questions (SKILL.md).
Boundary markers: There are no explicit delimiters or "ignore instructions" directives provided to ensure the agent distinguishes between its core coordination logic and potentially malicious content within messages or objectives.
Capability inventory: The skill enables high-privilege operations including arbitrary command execution (command), file modifications (write_file, apply_patch), and the ability to delegate tasks to other workers (delegate_task) (references/operations.md).
Sanitization: The documentation does not specify any sanitization, filtering, or validation requirements for the external data being processed.

horde-worker