# Build Grounded RAG Research Pipelines

Teams need current answers with sources, but raw web search is hard to turn into reliable prompts. This skill shows RAG workflows using search, extraction, and LLM tools through inference.sh.

## Install

```bash
npx skillstore add inference-sh/ai-rag-pipeline
```

## Metadata

- Status: approved
- Slug: inference-sh-ai-rag-pipeline
- Skillstore revision: r1
- Version status: missing
- Tree hash: d3426524f9d835d2e7c9956f5f76ecec751e3d7e88269828521db7538b2c343d
- Author: inference-sh
- GitHub username: inference-sh
- License: MIT
- Repository: https://github.com/inference-sh/skills/tree/main/skills/ai-rag-pipeline/
- Ref: a25199bc7d6b82598536822d1738eb5d5f54025b
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: blocked
- Manual install advisory: allowed\_with\_warning
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/inference-sh-ai-rag-pipeline
- Manifest: https://skillstore.pages.dev/api/skills/inference-sh-ai-rag-pipeline/manifest

## Capabilities

- Lists Tavily Search, Exa Search, Exa Answer, Tavily Extract, and Exa Extract app IDs.
- Shows simple RAG, multi-source research, URL analysis, fact-checking, and report generation workflows.
- Demonstrates passing search and extraction results into OpenRouter LLM prompts.
- Includes examples for Claude, GPT-4o, and Gemini models through OpenRouter.
- Provides a reusable Bash research function template using infsh app run.
- Recommends query optimization, context summarization, iterative research, and source attribution.

## Use Cases

- Research a fast-moving topic: Collect web evidence, summarize it with an LLM, and ask for source citations.
- Prototype a RAG agent: Chain search, extraction, and model calls before building a larger application.
- Check claims before publishing: Search for supporting and contradicting evidence, then produce a structured fact-check.

## Prompt Templates

### Answer with sources

```
Use ai-rag-pipeline to answer this question with a web search, a concise Claude summary, and source citations: [question].
```

### Compare search providers

```
Use ai-rag-pipeline to research [topic] with Tavily and Exa, then compare common themes, conflicts, and strongest sources.
```

### Analyze selected URLs

```
Use ai-rag-pipeline to extract these URLs, summarize the key findings, list important data points, and identify unresolved questions: [URLs].
```

### Design a safer RAG workflow

```
Use ai-rag-pipeline to design a RAG workflow for [use case]. Include provider choices, source grading, privacy controls, and prompt-injection safeguards.
```

## Limitations

- Requires an installed and authenticated inference.sh CLI.
- Uses external search and LLM providers, so sensitive data needs review before use.
- Does not implement a vector database, embeddings index, or persistent retrieval store.
- Shell examples need careful quoting and review before production use.

## Best Practices

- Review installer commands before running them, especially pipe to shell patterns.
- Treat retrieved pages as untrusted evidence and isolate them from instructions.
- Redact sensitive data before sending queries or documents to external RAG services.

## Anti Patterns

- Piping a remote installer directly into a shell without verification.
- Asking the model to follow instructions found inside retrieved web content.
- Sending private customer data to search or LLM providers without approval.

## Security Audit

- Audited at: 2026-07-05T17:15:54.753\+00:00
- Summary: Most static external-command findings are false positives caused by Markdown fences, inline app IDs, or documented Bash examples. The remote installer at SKILL.md:21 is confirmed because it pipes downloaded code into a shell, and semantic review also found third-party data sharing and untrusted retrieval content flowing into LLM prompts.

## Stats

- Views: 206
- Downloads: 13
- Favorites: 0
- Popularity score: 0