The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements logic to process, store, and retrieve user-supplied queries through caching layers (Redis and Qdrant). Untrusted data from user inputs is interpolated directly into prompts for downstream LLM API calls and stored in semantic caches without explicit sanitization or boundary markers.
Ingestion points: user_question and query parameters in functions like smart_llm_call, call_with_prompt_cache, and warm_cache within SKILL.md.
Boundary markers: Absent. Input is passed directly to LLM message structures or embedding models.
Capability inventory: The skill performs network operations (OpenAI/Anthropic API calls, local Redis/Qdrant connections) and local file reading (knowledge-base.txt).
Sanitization: No evidence of input validation, filtering, or escaping for the processed query strings.

llm-caching