Prompt dataset deduplication plan
For evaluation dataset curators: a duplicate and split-contamination report.
$12
Works with the AI Agents you already use
Prompt dataset deduplication plan
Example session with this skill installed
Use the skill to produce a duplicate and split-contamination report. Input S1: Train A: Reset my password. Test B: reset my password. Test C: Explain DNS caching.
- Read your context and instructions
- Compiled the prompt dataset deduplication
Synthetic test example. Recorded AI smoke-test response. No real service changes were executed.
A and B are a normalized duplicate across train and test; C is a separate topic in the supplied sample.
Comparison rule proposal: retain original IDs/text, then trim whitespace and lowercase for exact normalized comparison. Under that rule:
- Train A "Reset my password." -> "reset my password."
- Test B "reset my password." -> the same string.
- Test C "Explain DNS caching." -> "explain dns caching."
A and B are not byte-identical because capitalization differs, but are exact after this normalization. This contaminates the test's independence for the password-reset item. No broader semantic duplicate search was performed.
Proposed repair: keep A in training and mark B excluded from independent-test scoring, preserving its original record and a change ledger linking B to A. Alternatively move the entire password-reset family into one split before training; neither change was executed. Keep C in test provisionally, subject to family review of the full dataset.
Missing inputs: labels, full dataset, family definitions and desired split coverage.
Acceptance check: after approved changes, no normalized duplicate or related password-reset family crosses train/test, and all moved/excluded records retain provenance.
Connects securely to your tools. The creator never sees your data.
About this skill
Use this skill when you have a small prompt dataset with labels and split assignments and need a duplicate and split-contamination report.
The workflow:
- Normalize comparison features while retaining original text and IDs.
- Detect exact duplicates first, then label near-duplicates as review candidates.
- Keep semantically related families together when defining evaluation splits.
- Show every proposed merge or split move with its reason; do not silently delete records.
You receive the work product with input-linked findings, visible unknowns and concrete verification checks. Work from pasted or redacted text. This skill does not execute changes, send messages or access service accounts.
Original AI-assisted content. A synthetic worked example is included. Use through your Agensi-connected agent. No extra paid API or runtime is required for this workflow. Delivery is through the Agensi agent connection. No sales, ranking or compliance outcome is promised.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
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 buying a skill.
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 8 days ago
Permissions
No special permissions declared or detected