The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources including Railway project configurations, environment variables, and service listings which could potentially contain adversarial instructions designed to influence the agent.
Ingestion points: railway status, railway variables, and GraphQL queries against the Railway environment configuration (SKILL.md).
Boundary markers: The skill uses shell heredocs to prevent accidental command injection but does not define explicit prompt delimiters to isolate ingested data from instructions.
Capability inventory: The skill possesses the ability to execute shell commands via Bash(railway:*), stage and apply infrastructure changes, and delete services or volumes.
Sanitization: The skill explicitly instructs the agent to use heredoc syntax to prevent shell escaping issues during command execution, providing a layer of protection against traditional command injection.
[COMMAND_EXECUTION]: The skill relies on the execution of the railway-cli and a local vendor script railway-api.sh to perform its primary functions. This includes critical infrastructure operations such as creating new environments, switching context between environments, and performing mutations that can delete or modify live services.

railway-environment