skills.rest
HsunGong
HsunGongCOMMUNITY
Explore all skills by this author →
GitHub

generate-reasoning-traces

Generate Stage 2 reasoning traces with think blocks and generation plans.

Updated Jan 16, 2026
Search Tags:#generate#prompt-engineering#ai-research#datasets#SFT#reasoning-traces#stage-2
One-click install
npx skills add https://github.com/HsunGong/prep --skill generate-reasoning-traces
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: generate-reasoning-traces
Source: https://github.com/HsunGong/prep/tree/main/.github/skills/reasoning
Command: npx skills add https://github.com/HsunGong/prep --skill generate-reasoning-traces

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill automates the creation of Stage 2 reasoning traces, capturing user intent, inferred details, and generation plans to support transparency and auditability in AI reasoning workflows.

Core Features & Use Cases

  • Generate structured Stage 2 traces from prompts and user intents for AI research and model evaluation.
  • Produce deterministic, reusable traces in JSONL and human-readable summaries for documentation.
  • Use Case: When building SFT datasets, attach reasoning traces to prompts to improve interpretability and evaluation of multi-step reasoning.

Quick Start

Run the reasoning trace generator on a sample prompt using the included script. Example: python sft_generate_reasoning_traces.py --input prompts/sample.jsonl --output stage2_reasoning_traces/

Frequently Asked Questions about generate-reasoning-traces

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate reasoning traces for SFT dataset creation?▼

You generate reasoning traces for SFT datasets by running a Python script that processes input prompts in JSONL format and outputs structured Stage 2 traces with think blocks and generation plans for AI research.

What are Stage 2 reasoning traces in AI research?▼

Stage 2 reasoning traces are structured outputs capturing user intent, inferred details, and generation plans to support transparency and auditability in multi-step AI reasoning workflows and model evaluation.

Can I use this reasoning trace generator with JSONL prompt files?▼

Yes, you can use this generator with JSONL prompt files by passing the input path to the included script, which then produces deterministic reasoning traces and human-readable summaries in your specified output directory.

Do I need YAML frontmatter to configure reasoning trace generation?▼

Yes, YAML frontmatter is required to ensure consistent configurations during reasoning trace generation, guiding the structured output of think blocks and generation plans for your AI prompts.

What is the best way to document multi-step reasoning for model evaluation?▼

The best way to document multi-step reasoning is to generate deterministic, reusable traces in JSONL alongside human-readable summaries, attaching them to prompts to improve interpretability during model evaluation.

Related Skills

Explore More Education & Research Skills→
cas-bigdatalabcas-bigdatalab
541

tsv_formatter

Convert TSV files into JSONL format using Python.

Official
Basic
overviewlabsoverviewlabs

fine-tuning-with-trl

Fine-tune LLMs with TRL for SFT, DPO, and PPO/GRPO workflows.

Community
Advanced
ollieb89ollieb89

prompt-engineering-patterns

Create reusable few-shot and chain-of-thought prompt templates for LLM tasks.

Community
Advanced
cas-bigdatalabcas-bigdatalab
541

pdf_create

Generate PDF documents from structured JSON content with configurable page size and metadata.

Official
Intermediate