The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill incorporates an audit workflow in flow/review.md that uses a static analysis tool to scan project implementation files (e.g., lib/screens/checkout_screen.dart). This processing of untrusted project data constitutes an indirect prompt injection surface where malicious content within the audited code could attempt to influence the agent's behavior during the design review process.
Ingestion points: Automated analysis of source files within project directories during the review phase (flow/review.md).
Boundary markers: No specific delimiters or instructions to ignore embedded prompts are provided in the markdown procedures for file auditing.
Capability inventory: The skill executes commands through npx, spawns subprocesses for adb and xcrun, and runs headless Chrome for wireframe rendering.
Sanitization: There are no explicit content filtering or sanitization steps mentioned for the ingested codebase data.
[EXTERNAL_DOWNLOADS]: The skill requires the use of external packages from the npm registry via npx, specifically trunative (the author's companion tool) and @google/design.md. These resources are fetched and executed at runtime to perform diagnostics and specification linting.
[COMMAND_EXECUTION]: The skill relies on several shell commands for its core functionality, including environment validation (npx trunative doctor), static analysis (npx trunative detect), and mobile environment interactions using Android Debug Bridge (adb) and iOS simulator control (xcrun).

trunative-expo