The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides Retrieval-Augmented Generation (RAG) pipeline patterns that are susceptible to indirect prompt injection. 1. Ingestion points: The output of external search tools such as tavily/search-assistant and exa/search (as seen in the pipeline examples in SKILL.md) is used as primary context. 2. Boundary markers: The examples utilize simple text headers like 'Search Results:' or 'Source 1 (Tavily):' which provide minimal protection against adversarial instructions embedded in retrieved content. 3. Capability inventory: The skill facilitates the invocation of high-capability models (Claude 4.5, GPT-4o) using the belt app run command. 4. Sanitization: No sanitization, escaping, or validation steps are included in the provided shell script templates to handle potentially malicious content from external websites.
[EXTERNAL_DOWNLOADS]: The documentation includes a link to installation instructions and scripts hosted on GitHub (inference-sh/skills). This is a neutral reference to the infrastructure's official documentation for CLI setup.

ai-rag-pipeline