Works with the AI Agents you already use
Docwell RAG Pro — Agensi Skill Pack
A professional RAG toolkit for structured metadata extraction, pipeline design, and automated faithfulness auditing.
$6
Docwell RAG Pro — Agensi Skill Pack
Example session with this skill installed
I have this draft answer: "The student John Doe submitted his thesis on June 12th supervised by Dr. Smith." Use the provided passages to check for faithfulness and extract the metadata.
- Read your context and instructions
- Compiled the docwell rag pro
I have analyzed the draft answer against the provided document text and extracted the relevant metadata.
Document Metadata
| Field | Value | Source Page | Evidence | Confidence |
|---|---|---|---|---|
| student_name | John Doe | 1 | "Submitted by: John Doe" | high |
| submission_date | 2023-06-10 | 1 | "Date: June 10, 2023" | high |
| supervisor | Dr. Jane Smith | 1 | "Supervisor: Dr. Jane Smith" | high |
| document_type | Thesis | 1 | "A Senior Thesis submitted..." | high |
RAG Faithfulness Audit
- Verdict: Partially grounded.
- Claim Audit:
- "The student John Doe submitted his thesis": Supported. Source p.1 confirms name and doc type.
- "submitted... on June 12th": Contradicted. Source p.1 states June 10th.
- "supervised by Dr. Smith": Supported. Source p.1 confirms Dr. Jane Smith.
- Unsupported or conflicting details: The date in the draft (June 12th) conflicts with the document (June 10th).
Safer Answer
John Doe submitted his thesis on June 10, 2023, under the supervision of Dr. Jane Smith.
Next steps
- Update your database record to reflect the corrected submission date of June 10th.
- Verify if the "June 12th" date exists elsewhere in the document or if it was a generation error.
- Use the extracted JSON metadata to automate your indexing pipeline.
Connects securely to your tools. The creator never sees your data.
About this skill
PDF processing and Retrieval-Augmented Generation (RAG) workflows often suffer from "black box" behavior where sources are lost and hallucinations go unchecked. This skill pack provides a structured framework for extracting metadata, managing RAG logic, and auditing outputs for factual grounding.
What it does
- Metadata Extraction identifies and validates fields like authors, titles, and identifiers from PDF cover pages with confidence scores.
- Workflow Orchestration guides the design of RAG pipelines including chunking, embedding, vector indexing, and retrieval strategies.
- Faithfulness Auditing breaks down generated answers into atomic claims to verify them against source passages.
- Evidence Mapping attaches page numbers and source spans to every extracted data point or generated response.
- Grounded Correction produces revised answers that strip away unsupported claims found during the audit phase.
How it works
- Supply a document or retrieved text to extract structured metadata into JSON or table formats.
- Define RAG pipeline parameters such as chunk size, overlap, and top-k retrieval values based on the assistant's engineering guidance.
- Input a draft RAG answer alongside its retrieved passages to trigger a claim-by-claim verification.
- Review the groundedness verdict to identify unsupported names, dates, or quantities before deployment.
Frameworks & tools
This toolset is designed for developers building with vector databases like ChromaDB or Pinecone, and LLM frameworks such as LangChain or LlamaIndex. It focuses on the logic layer of PDF ingestion and retrieval verification.
Why this beats prompting it yourself
Generic prompts often fail to preserve strict evidence chains or ignore "null" values when data is missing. These skills enforce a rigorous verification procedure that treats model confidence as secondary to explicit source text, reducing the risk of silent failures in production RAG systems.
Use cases
- Automate the indexing of academic papers or corporate reports into a search database.
- Troubleshoot retrieval failures by auditing why specific claims in a RAG output are unsupported.
- Design a high-precision document Q&A system that cites page numbers for every sentence.
- Build an automated QA layer to flag and correct hallucinations in LLM-generated summaries.
Known limitations
Does not perform OCR, physical file uploads, or direct database writes. Requires the developer to provide the raw text or retrieved passages via an external tool or manual input.
Requires external tools for OCR and PDF parsing. Does not perform actual vector database operations or API calls. Verification is limited to the text provided in the context window.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed today
Requires