The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the project environment to determine framework behavior and architectural constraints, which could be exploited to influence the agent's actions if these files are compromised.
Ingestion points: The skill reads package.json, next.config.*, astro.config.*, and documentation files residing within node_modules/next/dist/docs/ (references/next.md).
Boundary markers: The instructions lack explicit delimiters or warnings to the agent to ignore potential instructions embedded within the data it reads from the project files or documentation.
Capability inventory: The agent is authorized to modify UI components, layout structures, and component architecture, providing a significant impact surface if redirected.
Sanitization: There is no evidence of validation or filtering for the content read from the project's dependency manifests or installed documentation.
[DYNAMIC_EXECUTION]: The skill incorporates dynamic discovery mechanisms to load additional instructions (skills) based on the presence of specific files or directories in the project.
Evidence: The skill instructs the agent to discover and load additional logic from the .agents/skills/ directory (e.g., 'Astro skill' or 'Next.js workflow skills') and specific utility skills like shadcn-component-boundaries (SKILL.md, references/astro.md, references/next.md). This represents dynamic loading of executable instructions from computed local paths.

project-architecture