qra

name	qra
description	Extract Question-Reasoning-Answer pairs from text. Use --context for domain-focused extraction. Validates answers are grounded in source text.
allowed-tools	Bash, Read
triggers	["extract QRA","extract Q&A","extract knowledge","create Q&A pairs","knowledge extraction","generate questions from"]
metadata	{"short-description":"Extract grounded Q&A pairs from text"}

QRA Skill

Extract Question-Reasoning-Answer pairs from text and store in memory.

Happy Path

# Extract from text file
./run.sh --file document.md --scope research

# With domain focus (recommended)
./run.sh --file notes.txt --scope project --context "security expert"

# Preview before storing
./run.sh --file transcript.txt --dry-run

# From stdin
cat meeting_notes.txt | ./run.sh --scope meetings

Parameters

Flag	Description
--file	Text or markdown file
--text	Raw text content
--scope	Memory scope (default: research)
--context	Domain focus, e.g. "ML researcher"
--dry-run	Preview without storing
--json	JSON output

What It Does

1. Split text into logical sections
2. Extract Q&A pairs via LLM (parallel batch)
3. Validate answers are grounded in source
4. Store to memory via memory-agent learn

When to Use

• Text content (not PDFs - use distill for PDFs)
• Meeting transcripts
• Code documentation
• Notes and summaries
• Any plain text you want to remember

Examples

# Meeting transcript
./run.sh --file meeting.txt --scope team --context "project manager"

# Code documentation
./run.sh --file README.md --scope code --context "Python developer"

# From clipboard/pipe
pbpaste | ./run.sh --scope notes --dry-run

Environment Variables (Optional Tuning)

Variable	Default	Description
QRA_CONCURRENCY	6	Parallel LLM requests
QRA_GROUNDING_THRESH	0.6	Grounding similarity threshold
QRA_NO_GROUNDING	-	Set to 1 to skip validation