The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A non-functional placeholder string sk-proj-abc123 is used in SKILL.md to illustrate insecure hardcoding of API keys. It is presented within a 'BAD' example block for instructional purposes and is not a functional secret.
[EXTERNAL_DOWNLOADS]: The skill's code snippets reference standard dependencies for LLM integration, including openai, @anthropic-ai/sdk, zod, and gpt-tokenizer.
[PROMPT_INJECTION]: The skill addresses indirect prompt injection surfaces by providing patterns for processing untrusted data. * Ingestion points: Untrusted data enters the agent context through the text parameter in extraction functions (e.g., extractContactOpenAI in SKILL.md) and the userMessage parameter in chatWithTools (references/advanced.md). * Boundary markers: The examples demonstrate the use of system messages to define agent behavior and separate it from user-provided content. * Capability inventory: The skill uses OpenAI and Anthropic API clients to process text and execute tool calls. * Sanitization: The code examples recommend the use of Zod for schema validation and parsing of structured outputs to ensure data integrity.

llm-integration