The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a template pattern that interpolates untrusted data directly into an LLM prompt, creating a surface for indirect prompt injection.
Ingestion points: The classifyTicketPrompt function in core/prompts/classify.ts takes a raw string input (ticket) and embeds it into the prompt.
Boundary markers: The prompt template lacks clear delimiters (like XML tags or triple quotes) or explicit system instructions to ignore embedded commands within the ticket content.
Capability inventory: The generated prompt is processed by the llmCall function, which performs network operations to the Anthropic API and executes JSON.parse on the response.
Sanitization: There is no evidence of input escaping, length limiting, or content filtering on the ticket variable before it is interpolated into the final prompt string.

llm-patterns