Skip to content
Back to skills

Flan Instruction Prompt Formatter

ASecurity

Render FLAN-style direct, few-shot, and chain-of-thought instruction prompt/completion pairs without leaking hidden answers into prompts.

  • 247 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 9, 2026
toolspython

Security analysis

A100/100

Scanned September 9, 2026

npx -y skills add VectorSpaceLab/AREX-Skill --skill flan_instruction_prompt_formatter --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Flan Instruction Prompt Formatter?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Flan Instruction Prompt Formatter
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vectorspacelab-flan-instruction-prompt-formatter/badge)](https://www.skillsdirectory.com/skills/vectorspacelab-flan-instruction-prompt-formatter)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

SKILL.md
---
name: flan_instruction_prompt_formatter
description: Render FLAN-style direct, few-shot, and chain-of-thought instruction prompt/completion pairs without leaking hidden answers into prompts.
---

# FLAN Instruction Prompt Formatter

Use this skill after mixture construction to format examples for instruction finetuning or held-out evaluation.

## Inputs

- Examples with `instruction`, `input`, `answer`, and optional `rationale`.
- Mode `direct`, `few_shot`, or `cot`.

## Outputs

- Prompt/completion records and metadata.

## Workflow

1. Build prompts from instruction and input fields.
2. Put target answers only in completions for query examples.
3. Put rationale plus final answer in CoT completions.
4. Record metadata for mechanism checks.

## Validation

Run `python tests/test_prompt_formatter.py`.

## Limitations

Uses compact deterministic templates rather than every original FLAN template.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…