The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by reading various project files to automate its setup.
Ingestion points: The skill scans the project root and subdirectories for content in files such as GEMINI.md, CLAUDE.md, .cursorrules, .github/copilot-instructions.md, and design-tokens.json to extract design direction and brand voice.
Boundary markers: The instructions do not define boundary markers or explicit prompts to ignore instructions embedded within the scanned files.
Capability inventory: The skill has the capability to write new component files, create a sink route, and execute shell commands via build tools and local scripts.
Sanitization: There is no mention of sanitization, validation, or filtering of the content extracted from the external project files.
[COMMAND_EXECUTION]: The skill executes various commands to fulfill its auditing and verification roles.
The agent is instructed to run node scripts/scan-components.js from the skill directory to generate inventory reports.
During the verification phase, the skill executes framework-specific build and lint commands such as pnpm build, pnpm lint, and hugo --minify.
[EXTERNAL_DOWNLOADS]: The skill suggests performing an accessibility audit using pnpm dlx @axe-core/cli. This uses a well-known accessibility auditing tool to verify the generated sink page.
[DYNAMIC_EXECUTION]: The skill uses a local script (scripts/scan-components.js) to dynamically analyze the project's component directory and generate reports based on the findings.

building-design-systems