The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions do not contain any patterns intended to bypass AI safety filters or override system instructions. Phrases like 'CRITICAL: Before Using This Skill' are used appropriately for developer guidance and documentation formatting.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill correctly advises against hardcoding sensitive information, specifically the TOGETHER_API_KEY, and recommends using environment variables via process.env. Network operations target the well-known Together AI API endpoint (api.together.xyz).
[UNVERIFIABLE_DEPENDENCIES]: The skill references standard, well-known packages such as together-ai, openai, and zod. These are established libraries for the described tasks.
[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for processing external data such as image URLs and text transcripts. It mitigates potential injection risks by recommending structured output formats (JSON schema) and explicit system prompt instructions to constrain model responses.
[DYNAMIC_EXECUTION]: Code examples demonstrate standard usage of JSON.parse for processing model outputs and Zod for schema generation. No unsafe dynamic execution patterns like eval() or exec() are present.
[PRIVILEGE_ESCALATION]: No commands for elevating privileges or modifying sensitive system files were found.
[PERSISTENCE_MECHANISMS]: The skill does not contain instructions to establish persistent access on the host system.

ai-infrastructure-together-ai