The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reviews application source code and project briefs (PRODUCT.md, DESIGN.md, STACK.md), which are untrusted ingestion points. These inputs could contain instructions designed to influence the agent's behavior. The skill grants the agent capabilities to write to the file system (briefs and wireframes) and execute shell commands, creating a pathway for exploitation through repository content.
Ingestion points: Source code files and project-level markdown briefs.
Boundary markers: None used to isolate ingested data from instructions.
Capability inventory: File system writing and shell command execution (npx, chrome, adb, xcrun).
Sanitization: No explicit sanitization or validation of the ingested repository content is performed before processing.
[COMMAND_EXECUTION]: The skill provides procedures for the AI agent to execute several CLI tools and shell commands to perform its design review role:
npx trunative commands (install, doctor, rubric, detect).
chrome --headless for generating wireframe images.
adb and xcrun simctl for mobile device screenshots and interactions.
[EXTERNAL_DOWNLOADS]: The skill relies on npx to fetch and execute external packages, specifically the vendor-provided trunative tool and the Google-owned @google/design.md package.

trunative-compose