The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It processes untrusted external data from web searches and user-provided URLs to guide its research and code generation phases.
Ingestion points: Target game links (repositories or game pages) and web research results (SKILL.md).
Boundary markers: Absent. The instructions lack explicit warnings to the agent to ignore instructions embedded in the external content.
Capability inventory: File system writes to the Games/ directory, shell command execution via bun and vite, and GitHub repository management including creating and merging pull requests.
Sanitization: Absent. External content is used directly to define game mechanics and systems without filtering.
[COMMAND_EXECUTION]: The skill performs standard development operations using the bun runtime.
Evidence: Use of bun run check-types, bun test, and vite for building and verifying the generated game slices.

harvest-game