The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest external data from GitHub pull requests using gh pr view and gh pr diff. This creates a surface for indirect prompt injection where malicious instructions embedded in a PR's description or code could attempt to influence the agent's review or testing actions.
Ingestion points: SKILL.md (Step 1: gh pr view, gh pr diff)
Boundary markers: Absent. The agent reads the raw output of these commands.
Capability inventory: Execution of build scripts (yarn build), native device commands (xcrun), and simulator navigation (agent-device).
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill relies on several shell-based tools to perform its functions, including git for branch management, gh for GitHub integration, yarn for project builds and testing, and xcrun for managing the iOS simulator environment.
[EXTERNAL_DOWNLOADS]: The skill performs local network operations using curl to check the status of the Metro bundler (http://localhost:8081) and fetch to send runtime debug data to a local server.
[DYNAMIC_EXECUTION]: In Step 6, the skill generates and runs an inline Node.js script using node -e to create a temporary HTTP server for logging debug values from the application to the filesystem.

review-and-test