Works with the AI tools you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIGitHub CopilotGitHub CopilotGemini CLIGemini CLI+20 more

    Prompt Datasets Builder

    2

    Professional curation skill for LLM fine-tuning datasets, deduplication, and leakage auditing.

    $5

    Secure checkout via Stripe

    30-day refund guarantee

    Converts to your local currency at checkout

    0 installsSecurity scanned
    Prompt Datasets Builder

    Prompt Datasets Builder

    Example session with this skill installed

    Review this set of 3 prompts for a customer support bot fine-tuning task. Check for duplicates, quality issues, and suggest a JSON schema for export.

    1. How do I reset my password?
    2. I need to change my password, how?
    3. Tell me a joke.
    • Read your context and instructions
    • Compiled the prompt datasets builder
    • Generated the document

    I have analyzed the provided prompts for duplication, relevance, and quality. I detected one near-duplicate pair and one out-of-scope record.

    Prompt Dataset Curation Report

    Dataset Inventory

    FieldTypePurposeMissingnessNotes
    raw_textStringUser intent0%Primary input

    Curation Findings

    IDRecordLabelRecommendationReason
    001How do I reset my password?High QualityKeepClear, specific intent.
    002I need to change my password, how?Near DuplicateMerge/RemoveSemantic overlap with 001.
    003Tell me a joke.Out of ScopeRemoveDoes not match support bot objective.

    Dataset Schema

    {
  "id": "string",
  "prompt": "string",
  "category": "account_management",
  "intent": "password_reset",
  "quality_score": "float",
  "metadata": {
    "source": "raw_input",
    "version": "1.0.0"
  }
}

    Safety & Sensitive Data

    No PII, secrets, or restricted content detected in the sample records.

    Next steps

    1. Consolidate records 001 and 002 into a single canonical "password_reset" instruction.
    2. Define a "support_functions" taxonomy to filter out generic conversational queries like jokes.
    3. Apply the provided JSON schema to your full dataset for export readiness.

    prompt-datasets-builder.pdf

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    What you get

    Deduplicate instruction-tuning sets to improve model training efficiency.Identify PII and secrets in prompt logs before fine-tuning.Detect train/eval leakage to ensure valid model evaluation metrics.Standardize inconsistent labels across large multi-source datasets.

    About this skill

    The problem

    Fine-tuning and evaluation pipelines fail when fed noisy, duplicate, or inconsistent prompt data. Manual curation is slow, error-prone, and often misses subtle train/eval leakage or safety risks.

    What it does

    • Identifies exact and near-duplicate prompt-response pairs to ensure dataset diversity.
    • Performs quality audits to flag ambiguous, incomplete, or contradictory records.
    • Detects data leakage between training and evaluation splits to prevent inflated performance metrics.
    • Scans for sensitive data, API keys, and credentials without reproducing them in the report.
    • Generates structured JSON/JSONL schemas and curation reports for immediate export readiness.

    Frameworks & tools

    Designed for LLM-ops workflows, supporting JSON, JSONL, CSV, and TSV formats for fine-tuning on platforms like OpenAI, Anthropic, or Hugging Face.

    Why this beats prompting it yourself

    Writing individual prompts to clean data often results in hallucinated examples or inconsistent labeling. This skill uses a multi-modal curation workflow that enforces strict provenance, prevents fabrication of metadata, and applies a systematic 23-point quality control process that simple prompts lack.

    Use cases

    • Cleaning raw instruction-tuning data before model training.
    • Auditing evaluation sets for leakage against training benchmarks.
    • Building consistent label taxonomies for multi-turn conversational datasets.
    • Redacting PII and secrets from proprietary prompt logs.

    Known limitations

    Does not execute dataset exports to external storage; it generates the structured content and reports for you to save. Cannot verify the factual accuracy of responses without external ground-truth data.

    How to install

    Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.

    ~30 seconds
    1. 1

      Download the ZIP

      Free skills download straight away. Paid skills unlock right after purchase.

    2. 2

      Unzip into your skills folder

      Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.

    3. 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 6 days ago

    • Passed all security checks, Safe to install

    Listed6 days ago

    What's inside

    Frequently Asked Questions