The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents how to handle untrusted data at project boundaries, which naturally creates a surface for indirect prompt injection attacks where data could contain instructions.\n
Ingestion points: The skill describes processing API contracts, request bodies, and environment variables as documented in SKILL.md and references/setup-core.md.\n
Boundary markers: Zod schemas are used as the primary structural boundary; however, specific natural language delimiters for prompt isolation are not mentioned.\n
Capability inventory: The skill includes standard package management instructions via bun but no scripts that execute arbitrary commands.\n
Sanitization: Zod's intrinsic function is to validate and sanitize structural data.\n- [DYNAMIC_EXECUTION]: The skill explains the z.compile() feature in Zod v4.5+, which uses new Function to optimize schema validation performance.\n
Evidence: Detailed in references/setup-core.md under the AOT compilation section.\n
Note: The instructions correctly advise on using jitless: true to disable this behavior in restricted environments.\n- [EXTERNAL_DOWNLOADS]: The skill provides references to official and trusted documentation for the Zod library.\n
Evidence: Mentions zod.dev and github.com/colinhacks/zod in references/source-map.md.\n
Note: These are official resources for the library described and do not represent a security risk.

zod