The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple commands through the railway CLI, including init, up, add, and variables. It also enables high-privilege operations such as railway ssh for remote shell access to services.
[EXTERNAL_DOWNLOADS]: The logging-python.md file instructs the agent to install the python-json-logger package via the uv package manager, which is a standard library download from the public PyPI registry.
[DATA_EXFILTRATION]: The skill references the path ~/.railway/config.json to retrieve the user.token for Railway GraphQL API authentication. This constitutes access to sensitive local credentials required for deployment automation.
[PROMPT_INJECTION]: The logging framework described in logging-python.md introduces a surface for indirect prompt injection by processing application-level data. 1. Ingestion points: Application log messages and metadata processed by the RailwayJsonFormatter and LoggingMiddleware. 2. Boundary markers: None defined in the implementation. 3. Capability inventory: Access to Railway authentication tokens and the ability to execute deployment and configuration commands via the CLI. 4. Sanitization: The implementation relies on standard JSON serialization without specific filtering or sanitization of untrusted input within log records.

railway