Works with the AI Agents you already use
AI Output Quality Guardian
Catch unsupported, incomplete, inaccurate, or unusable AI answers with a repeatable evidence-aware review and release gate.
$9.99
AI Output Quality Guardian
Example session with this skill installed
Review this assistant response before delivery. User asked for the current refund policy and a calculation for a $240 order with a 15% restocking fee. Draft answer: “Your refund is guaranteed within 24 hours under our latest 2026 policy. The fee is $24, so you will receive $216. I checked your account and submitted the refund.” No policy document, account access, or payment tool result is available. Identify blockers, evaluate each claim and calculation, repair only what can be safely stated, and provide a release decision.
- Read your context and instructions
- Compiled the ai output quality
Pre-Delivery Quality Review
Release decision: BLOCK
The draft makes current-policy, timing, account-access, and refund-completion claims without evidence. Do not send it as written.
Acceptance criteria
| Criterion | Result | Evidence / action |
|---|---|---|
| Current policy is sourced | Fail | No policy document or current source was provided. Remove the “latest 2026 policy” claim. |
| Refund timing is supported | Fail | “Guaranteed within 24 hours” is unverified; do not promise it. |
| Fee calculation is correct | Fail | 15% of $240 is $36, not $24. |
| Net amount is correct | Fail | If the fee applies to the full order and no other deductions apply, $240 − $36 = $204. State this as a conditional calculation, not a confirmed refund amount. |
| Account was checked | Fail | No account access is available; do not claim inspection. |
| Refund was submitted | Fail | No payment tool result exists; do not claim action completion. |
Safe repair
“I can’t verify the current refund policy or your account from the information available. If the 15% restocking fee applies to the full $240 order and there are no other deductions, the fee would be $36 and the remaining amount would be $204. Please check the current policy and confirm whether the fee applies before relying on that estimate. No refund has been submitted.”
Verification still needed
Obtain the current policy source, confirm the fee basis and any exclusions, and inspect the authorized account/payment system if the user wants a refund action. If a refund is later submitted, confirm the provider result and resulting transaction state before saying it is complete.
Residual uncertainty
The policy, eligibility, fee base, other deductions, timing, and account state remain unknown. The arithmetic is verified, but the conditional net amount is not a guaranteed refund quote.
Connects securely to your tools. The creator never sees your data.
About this skill
AI Output Quality Guardian
Review finished AI-generated answers and deliverables before release. Apply a clear rubric to the user's objective, required content, factual support, completeness, format, usability, safety, privacy, and risk. Identify material defects, repair them when authorized, recheck the result, and report any remaining uncertainty.
What it evaluates
- Instruction and format compliance against the actual request.
- Factual claims, sources, citations, calculations, freshness, and uncertainty.
- Completeness, contradictions, missing edge cases, relevance, and actionability.
- Code, data, JSON, links, files, and tool results when verification capabilities are available.
- Safety, privacy, localization, professional quality, and whether claimed actions have evidence.
How it works
Convert the request into acceptance criteria. Inspect the deliverable and supporting evidence. Score each criterion as pass, partial, fail, or unverified with specific evidence. Verify material claims when tools permit. Repair defects surgically when editing is allowed. Recheck the repaired output and issue a release decision with blockers, fixes, and residual risks. If no rubric exists, derive concise criteria and label them as inferred.
Principles
Judge observable output, not private chain-of-thought. Correctness outranks polish; fit-for-purpose quality is more useful than unnecessary length. Never invent facts, sources, citations, test results, files, credentials, or completion. Current claims need appropriately fresh evidence. Stronger checks are required for high-impact content. Preserve good work and the user's objective.
Boundaries
This skill audits final outputs and artifacts using available evidence. It does not request or expose hidden reasoning, guarantee correctness, or claim verification beyond checks actually performed.
Quality judgments depend on the user requirements and available evidence. Current facts, high impact claims, citations, code, and external actions require suitable sources or tests. Review can reduce defects but cannot guarantee correctness or completeness; unperformed checks remain unverified.
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 1 month ago
Permissions
Use file access only for user-authorized candidate outputs and supporting evidence. Edit only when requested or clearly authorized; preserve originals where practical. Do not claim tests, citations, source checks, or artifact validation unless actually performed. Never request hidden reasoning or expose sensitive data in review notes.