The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted GitHub issue content (titles, bodies, comments, and metadata) to guide the debugging and fixing workflow.
Ingestion points: External data from GitHub issues is used to define the agent's tasks in SKILL.md (e.g., "Understand the issue and reproduce the problem").
Boundary markers: The skill contains a specific "Security" section warning the agent that issue content is untrusted and explicitly instructs it to ignore embedded instructions that attempt to bypass safety filters or exfiltrate data.
Capability inventory: The agent has access to shell execution (yarn, adb, xcrun, lsof, kill), file system read/write, and the ability to raise Pull Requests.
Sanitization: No programmatic sanitization or strict schema validation is applied to the issue data before processing; the skill relies on instructional guardrails.
[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands for building, testing, and environment cleanup.
Process Management: Uses lsof and kill -9 to terminate background processes (specifically Metro bundler on port 8081) during the cleanup phase.
Mobile Tooling: Uses adb for Android emulator configuration and xcrun simctl for iOS simulator state checks.
Build Tooling: Executes yarn commands for building the fixture app, running E2E tests (detox), and starting the development server.

fix-github-issue