dataset-synthesizer-revisor

What problem does it solve?

This Skill prevents low-quality JSONL fine-tuning datasets from being used for training by detecting placeholders, AI apology traces, JSON syntax errors, and content imbalance that would degrade model behavior.

Core Features & Use Cases

• Placeholder and marker detection: Identifies unresolved tokens such as [Preencher...], [INSIRA...], [Seu nome] and other generator artifacts.
• AI-error and apology tracing: Flags machine-error remnants and "AI-splaining" phrases that should not appear in production training data.
• JSONL structural validation: Verifies each line is a standalone valid JSON record with required root keys and correct escaping to avoid ingestion failures.
• Line-by-line audit reporting: Produces a Markdown report with total lines analyzed, problematic line counts, per-line anomaly descriptions, and precise remediation recommendations saved next to the original dataset.
• Use Case: Data QA and ML engineers validating synthesized datasets for customer support or documentation fine-tuning workflows.

Quick Start

Use the skill to analyze the file path/to/dataset.jsonl and produce a line-by-line Markdown audit report saved next to the original file.