The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill suggests that the agent execute the xcobra package using bunx (in references/upgrading/new-architecture.md) to verify the state of the React Native New Architecture. The xcobra package is not an official tool from Expo or a recognized trusted vendor, and its execution via bunx facilitates the download and execution of unverified third-party code.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to install dependencies (npm install) and execute local Node.js scripts (scripts/fetch.js, scripts/validate.js) as well as various CLI tools such as eas, expo, and bunx to manage builds and deployments.
[EXTERNAL_DOWNLOADS]: The skill includes a custom utility (scripts/fetch.js) designed to download content from remote URLs. While the documented use cases involve fetching official Expo schemas and documentation, the script itself does not implement URL restrictions or domain whitelisting, allowing the agent to fetch content from arbitrary external sources.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where the agent fetches and processes external documentation and schemas to determine its subsequent actions. This creates an attack surface for indirect prompt injection if the remote resources are compromised.
Ingestion points: External documentation and JSON schemas retrieved via scripts/fetch.js, and local workflow files (.eas/workflows/*.yml) processed by scripts/validate.js.
Boundary markers: The instructions do not specify any boundary markers or delimiters to isolate untrusted external content from the agent's core instructions.
Capability inventory: The skill has significant capabilities including network requests, file system access (via Node.js scripts), and the ability to execute shell commands and build tools.
Sanitization: There is no evidence of sanitization or strict validation of the content retrieved from external URLs before it is processed by the agent.

expo