The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides an installation method that fetches and executes a shell script from the official Railway installation endpoint and includes a command for updating the tool.
Evidence: The instruction bash <(curl -fsSL cli.new) in SKILL.md and the railway upgrade command both involve executing code downloaded from remote sources.
[PERSISTENCE]: The reference guide contains commands to append shell completions to configuration files like .bashrc and .zshrc.
Evidence: Found in reference.md under the railway completion command section: railway completion bash >> ~/.bashrc.
[COMMAND_EXECUTION]: Commands such as railway run and railway shell allow for the execution of arbitrary local processes with injected secrets.
Evidence: These commands are intended to run migrations or scripts with production/staging credentials injected into the environment.
[INDIRECT_PROMPT_INJECTION]: The skill's ability to run commands based on project state creates a surface for executing instructions found in untrusted data.
Ingestion points: Local project files, environment configurations, and command arguments supplied during runtime.
Boundary markers: No explicit boundary markers or warnings are provided to the agent to distinguish between instructions and data.
Capability inventory: Shell command execution, file system access, and environment variable management.
Sanitization: No sanitization or validation of the commands passed to the railway run utility is described.
[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to download the CLI tool from various package registries.
Evidence: Installation methods include brew install railway and npm i -g @railway/cli.

railway-cli