How to use this guide

This page explains what you need, how to get the files, how pages are laid out, and how to read prompts and scripts.

What you need

Get the repository

The repository holds the scripts and the sample data. Copy it once, then work from its folder in a terminal.

Copy the repository to your computer:

git clone https://github.com/GSA/rapid_upskill.git

Then move into the new folder:

cd rapid_upskill

Then check your Python version. The number after Python must be 3.10 or higher:

python3 --version

Run every command in this guide from this folder, called the repository root.

Reading paths

Goal Read
New to this guide: the framework’s shape, what you need, and the running example Getting started
See all five stages on one page Pipeline overview
Follow one small sample program The running example
Learn where people decide and how to batch work Human roles, gates and batching
Check what your platform must offer Platform requirements
Work through Stage 1 with sample prompts and scripts Stage 1 Knowledge acquisition
Work through Stage 2 with sample prompts and scripts Stage 2 Content development
Work through Stage 3 with sample prompts and scripts Stage 3 Review and verification
Work through Stage 4 with sample prompts Stage 4 AI-tutor coaching
Work through Stage 5 with sample prompts and scripts Stage 5 Assessment development
See what sits alongside or after the five stages Beyond the five stages
See the parallel certification-alignment workstream Certification alignment
See how delivery tooling turns chapters into slides, infographics, and finished volumes Delivery
Learn the cross-cutting operating practices behind every stage Operating practices
Follow the running example through every stage in one thread Worked example, start to finish
See what evidence actually backs this guide’s own claims Evidence and limitations
Look up a term Glossary
Adapt this method to a new domain or platform Extending this guide

Anatomy of a stage page

A stage page is a page for a sub-stage or a stage of the framework. All five stages are now published. Each stage page has these sections, in this order.

  • Outcome: what you will have at the end.
  • Where it fits: what comes before and after, and what this part takes in and hands on.
  • Why this way: the reasons for the approach, with a source for every number.
  • Steps: numbered actions, each starting with a verb.
  • Artifacts and formats: the files the steps produce, and their formats.
  • Prompts: links to the prompts the page uses.
  • Scripts: links to the scripts the page uses.
  • Definition of done: conditions you can check.
  • Common failures: what goes wrong, and how to spot it.
  • Adapting to your platform: what the steps need, and how to supply it elsewhere.
  • Where humans decide: each point where a person must decide or approve.

The exact template is in the authoring conventions.

Prompts

A prompt is one file of instructions for your platform. A placeholder is a name in double curly braces, such as {{OBJECTIVE_ID}}. Replace each with your own value before you run the prompt. The capabilities line lists the capabilities your platform must offer, for example file-read or web-fetch.

Scripts

Before you let a script write files, run it as a dry run. By this guide’s convention, a script that writes or deletes files does nothing until you pass --write. Without that flag, it reports what it would change.

A script that calls a language model uses the offline mock model provider by default. It needs no key. It stands in for a real model and is not a dry run. From the repository root, try it. The -B flag stops Python from writing cache folders.

python3 -B scripts/common/llm_adapter.py --prompt "Hello"

For the prompt Hello, the command prints this one line:

MOCK:185f8db32271

The mock provider gives the same reply every time for the same prompt. A different prompt gives a different reply. On Python older than 3.10, the tool prints one line and stops with exit code 2.

The guided hands-on task is Your first prompt. It takes about 15 minutes. You add a practice prompt, run the checker, and remove the practice files. Nothing is sent to an AI model.

Status words and IDs

A page’s status is draft (still being written), reviewed (a second person or agent has checked it against sources), or stable (reviewed, and also read on the live site). Prompts, scripts, and sub-stages have IDs such as P-S1-04, X-OP-01, and S1.4a. The codes CA (certification alignment), DL (delivery), and OP (operating practices) do not label stages. Certification alignment is a parallel workstream, delivery is outside the framework, and operating practices apply across stages. The authoring conventions cover both.

Scope of this guide

This guide is platform-neutral. It describes a design and how the reference implementation was run, as of September 2026. It makes no claims about results.

Terms

Pages explain each term where it is introduced. The Glossary gathers the key terms in one place.

Next: The running example.


To the extent possible under law, copyright and related rights in this work are waived under CC0 1.0 Universal.

This site uses Just the Docs, a documentation theme for Jekyll.