The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of agents that ingest untrusted data (free-form user input and uploaded files) while possessing powerful capabilities such as code execution in a sandbox, database manipulation, and email dispatch.
Ingestion points: Data enters the agent context via input_json (free-form text or JSON) and file reading through tools.app_files, as described in SKILL.md and references/manifest-tools.md.
Boundary markers: The skill documentation recommends using the system prompt as a strict input contract and highlights the use of an ephemeral code sandbox to isolate file processing tasks.
Capability inventory: As detailed in SKILL.md and the tool reference, agents can execute arbitrary code in a sandbox, perform database writes (tools.app_data_write), send emails (tools.email), and interact with third-party SaaS services through connectors.
Sanitization: The skill relies on the platform's manifest-level authority checks and sandbox isolation; it does not mention specific procedures for sanitizing or escaping untrusted content before processing.
[EXTERNAL_DOWNLOADS]: Fetches example projects, manifests, and configuration from the official Railcode-HQ GitHub repository using curl and tar.
[COMMAND_EXECUTION]: Instructs the user to install the railcode CLI globally via npm and use it to test and manage agents, including running manifests and managing schedules.

create-railcode-agent