The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external lesson content from JSON files which represents a potential attack surface for indirect prompt injection. However, it implements robust validation to mitigate this risk.
Ingestion points: The validate-lessons.ts script in references/implementation-guide.md reads content from all .json files in the src/lessons directory.
Boundary markers: The skill employs a strict Zod schema (LessonSchema) to define and enforce expected data structures for lesson IDs, titles, languages, levels, and phrase components.
Capability inventory: The skill uses fetch for API communication and npm test for test execution against the Speak API.
Sanitization: External data is sanitized via LessonSchema.safeParse, which ensures files conform to the expected schema before processing, rejecting invalid or unexpected structures.
[EXTERNAL_DOWNLOADS]: The skill configuration utilizes external resources and dependencies for CI/CD integration and testing.
Evidence: Fetches the @speak/language-sdk, zod, and vitest packages via npm as part of the implementation and testing workflow.
Evidence: References official GitHub Actions (actions/checkout@v4, actions/setup-node@v4) for workflow orchestration in the CI pipeline files.

speak-ci-integration