The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an 'AUDIT' mode that processes existing codebases, creating a surface where malicious instructions embedded in the analyzed code could influence agent behavior.
Ingestion points: The AUDIT mode described in SKILL.md takes an existing codebase as input for review.
Boundary markers: There are no specific instructions or delimiters provided to the agent to treat the input codebase as untrusted or to ignore embedded instructions.
Capability inventory: The skill references several powerful shell commands in references/dev-loop.md (e.g., pod install, npx expo run:ios, npx expo export), which the agent might be prompted to execute.
Sanitization: No sanitization or validation mechanisms are mentioned for the codebase data being processed.
[COMMAND_EXECUTION]: The documentation includes instructions for executing various development and build commands, which are standard for the mobile development workflow but involve shell access.
Evidence: references/dev-loop.md lists commands like pod install, npx expo run:ios, and npx expo export for native builds and project management.

mobile-architecture