The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a specific binary located at ${CLAUDE_PLUGIN_ROOT}/bin/helix (or helix.cmd on Windows) to perform rail management tasks. This execution is explicitly defined and limited via the allowed-tools configuration, ensuring the agent only uses the approved toolset.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest instructions from external 'rail' files that define rules and procedures the agent must follow.
Ingestion points: External rail files stored in ~/.helix/rails and the project directory are read via the helix rails for and helix rails get commands.
Boundary markers: The skill does not explicitly define delimiters for the ingested content, though the instructions state that the agent should treat the output as binding standards.
Capability inventory: The skill can execute the helix binary via Bash or PowerShell and perform standard project file modifications.
Sanitization: No specific sanitization or validation logic for the content of the rail files is mentioned.

rails