The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute unverified external code via package runners.
In references/new-architecture.md, it suggests running bunx xcobra expo eval "_IS_FABRIC". The package xcobra is an unknown third-party tool not part of the official Expo or React Native ecosystems.
In SKILL.md, it instructs the use of npx --yes submit-expo-feedback@latest. Executing unversioned code from an external source via npx is a high-risk pattern, especially for a tool not associated with an official vendor.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands for project maintenance, including recursive file deletion (rm -rf node_modules .expo) and native project regeneration (npx expo prebuild --clean). While these are standard developer tasks, they provide a powerful capability set that could be abused if paired with malicious input.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection.
Ingestion points: The skill reads local configuration files (package.json, app.json, AGENTS.md) and instructs the agent to consult external release notes at expo.dev/changelog.
Boundary markers: None are defined to distinguish between project data and agent instructions.
Capability inventory: The skill has access to shell execution (npx, bunx, pod, gradlew) and file system manipulation (rm).
Sanitization: There is no evidence of sanitization or validation of the content read from external documentation or local files before it influences agent actions.

expo-upgrade