Works with the AI tools you already use
Prompt Dataset Generator
Generate, clean, and augment structured prompt datasets for LLM training and evaluation.
$6.99
Prompt Dataset Generator
Example session with this skill installed
Create 5 instruction-tuning records in JSONL for a customer support agent. Categories: billing, technical, account. Vary difficulty from easy to hard.
- Read your context and instructions
- Compiled the prompt dataset generator
I have generated 5 diverse instruction-tuning records for customer support, covering billing, technical, and account categories across varying difficulty levels.
Customer Support Dataset
{"id": "CS_001", "instruction": "Explain the late fee policy to a customer.", "input": "Customer: Why was I charged an extra $15 this month?", "output": "Our policy includes a $15 late fee for payments received more than 3 days after the due date. You can avoid this by setting up autopay in your settings.", "category": "billing", "difficulty": "easy"}
{"id": "CS_002", "instruction": "Troubleshoot a synchronization error between the mobile app and desktop client.", "input": "User reports that their notes are not appearing on their phone after saving on their PC.", "output": "First, verify both devices are on the same version. Then, check if 'Cloud Sync' is toggled on in the mobile settings under 'Account > Data'. If the issue persists, clear the app cache.", "category": "technical", "difficulty": "hard"}
{"id": "CS_003", "instruction": "Process a request to update an email address.", "input": "I need to change my login email from old@mail.com to new@mail.com.", "output": "To update your email, go to Profile > Security > Update Email. You will receive a verification link at your new address to confirm the change.", "category": "account", "difficulty": "easy"}
{"id": "CS_004", "instruction": "Handle a disputed charge where the customer claims they already cancelled.", "input": "I cancelled my trial last week but was still billed for a full month.", "output": "I apologize for the oversight. I have verified your cancellation date of Oct 14th. I am processing a full refund for the charge, which should appear in 3-5 business days.", "category": "billing", "difficulty": "medium"}
{"id": "CS_005", "instruction": "Diagnose a recurring login loop issue.", "input": "Every time I enter my 2FA code, the page just refreshes back to the login screen.", "output": "This usually indicates a browser cookie conflict. Please clear your browser cookies for our domain or try opening the site in an incognito window to bypass the loop.", "category": "technical", "difficulty": "medium"}
Quality Summary
- Records processed: 5 created.
- Categories covered: billing (2), technical (2), account (1).
- Duplicate status: 0 duplicates identified; high lexical diversity.
- Schema status: Validated JSONL format.
Next steps
- Review the hard technical record for domain-specific accuracy.
- Export the JSONL to your local environment for model fine-tuning.
- Request 50+ additional records to improve class balance for the 'account' category.
Connects securely to your tools. The creator never sees your data.
What you get
About this skill
The problem
Manually creating diverse datasets for model fine-tuning or evaluation is slow and prone to repetition. Developers often end up with imbalanced labels, inconsistent schemas, or near-duplicate records that degrade model performance.
What it does
- Generates structured datasets for instruction tuning, classification, and evaluation.
- Augments existing seed examples by varying wording and context while preserving intent.
- Cleans up raw data by identifying duplicates, repairing schemas, and flagging ambiguities.
- Performs quality checks on label consistency, class balance, and JSON/JSONL structure.
Frameworks & tools
Generates outputs compatible with Hugging Face Datasets, PyTorch, and TensorFlow. Supports JSON, JSONL, and custom CSV/Markdown schemas.
Why this beats prompting it yourself
Standard LLM prompts often produce repetitive or "hallucinated" data structures. This skill enforces strict schema validation and applies specific diversification modes to ensure your training data covers edge cases and maintains class balance without manual oversight.
Use cases
- Generate 100+ instruction-tuning pairs for a specific domain like legal or medical.
- Build a balanced sentiment analysis dataset with specific custom labels.
- Create an evaluation suite to test a model's failure modes in complex reasoning.
- Normalize and de-duplicate a messy collection of scraped prompt ideas.
Known limitations
Cannot provide statistical representativeness or quality metrics without external evidence. Does not support generation of malicious or unauthorized security datasets.
How to install
Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.
- 1
Download the ZIP
Free skills download straight away. Paid skills unlock right after purchase.
- 2
Unzip into your skills folder
Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.
- 3
Ask your agent to use it
Restart the agent if it was already running. It picks the skill up automatically - no config needed.
Skills folder by agent
Click the path to copy it. Create the folder if it does not exist yet.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
30-day refund
Not a fit? Get your money back
Trust & safety
Security scanned
Verified clean 9 days ago
- Passed all security checks, Safe to install