- Home
- Skills
- Prompt & Skill Engineering
- VERUM Output Reviewer
Works with the AI Agents you already use
VERUM Output Reviewer
Evidence-based QA skill to flag factual drift, unsupported claims, and instruction violations in AI-generated responses.
Free
VERUM Output Reviewer
Example session with this skill installed
Review this response.
Ref: [ID-1] The fee is $50 for adults and $25 for kids.
Task: List fees.
Response: The fee is $50 for adults. Kids are free this month.
- Read your context and instructions
- Compiled the verum output reviewer
The review is complete. The candidate response failed to adhere to the reference material regarding child pricing and introduced an unsupported claim about a promotion.
VERUM Review Report
| ID | Candidate Passage | Reference/Instruction | Issue | Severity |
|---|---|---|---|---|
| 1 | "Kids are free this month." | [ID-1] The fee is... $25 for kids. | Direct contradiction of pricing and unsupported promotional claim. | SUBSTANTIAL |
| 2 | Omitted $25 fee. | Task: List fees. | Failed to list all applicable fees defined in source. | MINOR |
Summary
The response contains a factual contradiction regarding child pricing and includes an invented promotional offer not present in the reference material. The overall status is FAIL due to the substantial factual error.
Decision: FAIL
Rationales
- The claim that kids are "free" contradicts the explicit $25 fee in [ID-1].
- The "this month" constraint is an unsupported addition not found in the source text.
Corrected Response
The fee is $50 for adults and $25 for children.
Next steps
- Update the generation prompt to strictly prohibit promotional inventions.
- Verify if the "free" promotion exists in other source documents not provided here.
- Re-run the review if additional context regarding seasonal discounts is available.
Connects securely to your tools. The creator never sees your data.
About this skill
LLMs often hallucinate facts, drift from original instructions, or include unsupported claims that look plausible but lack evidence. Verum Output Reviewer provides a structured QA framework to audit AI-generated content against specific reference documents and mandatory constraints.I t reduces ambiguity in manual review by forcing a rigorous comparison between the original prompt, the source material, and the candidate response.
What it does 🔘 Factual Audit: identifies contradictions between the AI response and provided source materials using specific reference IDs. 🔘 Instruction Adherence: flags violations of mandatory conditions, formatting requirements, or tone constraints set in the original prompt. 🔘 Support Validation: detects "unsupported additions" where the AI adds information not found in the reference material. 🔘 Structured Feedback: delivers evidence-backed annotations with severity levels including Substantial, Minor, or Not Verifiable. 🔘 Decision Logic: applies a strict precedence-based grading system: Pass, Needs Revision, Not Verifiable, or Fail.
How it works
- Input Intake ➡️ accepts the original instruction, reference material, and the candidate AI response as three distinct blocks.
- Analysis ➡️ maps candidate passages against source IDs and mandatory requirements to identify drift or contradictions.
- Evaluation ➡️ applies a 1-5 rubric to language quality, consistency, and usefulness based solely on the provided context.
- Reporting ➡️ generates a structured review template containing finding rationales and, optionally, a corrected version of the text.
Frameworks & tools This skill is framework-agnostic and operates on raw text. It is designed for RAG (Retrieval-Augmented Generation) pipelines, content QA workflows, and internal training material audits.
Why use a structured reviewer Generic “check this for errors” prompts can produce inconsistent evaluations because the review criteria are not explicitly defined. VERUM applies a documented rubric and decision hierarchy, requiring findings to be tied to the supplied instruction or reference material rather than relying on unsupported assumptions.
Use cases
- RAG QA: verifying that an agent's summary accurately reflects the retrieved documentation chunks.
- Compliance Review: Policy & Guideline Review — checking whether AI-generated content preserves mandatory requirements contained in supplied policies, approved wording, or style guides.
- Technical Documentation Audit: ensuring training summaries don't invent numbers or units missing from the source.
- Response Refinement: identifying why a model failed a specific instruction and generating a corrected alternative.
Known limitations This skill is not a live web fact-checker or search engine. It cannot verify truth "in the world," only consistency against the provided text. It does not compute confidence percentages.
Cannot perform external web searches. Limited to supplied text only. Does not calculate mathematical averages or confidence scores. Requires three distinct inputs for full verification.
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
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 27 days ago
- Free with an Agensi account
Permissions
No special permissions declared or detected