The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill demonstrates secure practices by using environment variables (e.g., process.env.OPENAI_API_KEY, process.env.ANTHROPIC_API_KEY) for authentication rather than hardcoding sensitive credentials.
[EXTERNAL_DOWNLOADS]: The skill references and utilizes official SDKs from well-known and trusted AI service providers, including OpenAI and Anthropic, for its integration patterns.
[PROMPT_INJECTION]: The RAG (Retrieval-Augmented Generation) pipeline example in SKILL.md contains an architectural surface for indirect prompt injection common in such implementations.
Ingestion points: The ragQuery function ingests untrusted user input via the question variable and retrieved document content via the context variable.
Boundary markers: The implementation uses simple text labels (Question:, Answer based on this context:) but lacks robust delimiters or explicit instructions to the model to ignore embedded commands within the retrieved documents.
Capability inventory: The skill uses the openai.chat.completions.create and anthropic.messages.create tools to process these interpolated strings.
Sanitization: No sanitization or validation logic for the external context or user input is present in the provided code snippets.

llm-application-dev