100% Open Source100% Free100% Local

Babysit AIAutonomous AI

Make the important decisions, let AI do the rest.

Try:

copy

$ npx @gemstack/the-frameworkpnpm dlx @gemstack/the-frameworkbunx @gemstack/the-frameworkyarn dlx @gemstack/the-framework# One-shot (no install)

Or install:$ npm i -g @gemstack/the-frameworkpnpm add -g @gemstack/the-frameworkbun add -g @gemstack/the-frameworknpm i -g @gemstack/the-frameworkcopy

What is it?

It turns AI agents into autonomous teammates that handle work end-to-end — while you stay in control of key decisions.

Any software

(Semi-)autonomously build anything from simple web apps to complex software.

On this page

Stop babysittingAutonomous AIHow it worksFeaturesYour framework

Stop babysitting

Problem

AI is lazy

Bad fix

Appending "DON'T BE LAZY" to your prompts → minimal improvement.

Solution

Divide-and-conquer: The Framework instructs AI to split large tasks into smaller subtasks. By focusing on one unit of work at a time, AI spends significantly more effort overall, resulting in much higher-quality output.

Solution

Coverage guarantees: The Framework lets AI enumerate everything that needs to be done before writing code. AI then works through that checklist, ensuring comprehensive coverage and preventing lazy shortcuts.

Problem

Lazy AI plans

Bad fix

Explicitly tell AI to deep dive important aspects.

Solution

Automatic loop of critcal feedback ↔ research ↔ confidence ↔ implementation.

Problem

Lazy low-quality code

Despite AI being able to write high-quality code, it often goes for the quickest/laziest solution (not correct, not maintainable, not DRY).

Bad fix

Appending "WRITE CLEAN CODE" to your prompts → minimal improvement.

Solution

When an agent makes complex changes, the agent adds post-merge refactoring prompts to the AI Queue (with low priority).

Solution

Automatic routine security and code quality prompts (when your usage quota has plenty of capacity).

Problem

AI makes important decisions without asking

Bad fix

Tell AI "don't do this, research alternatives".

Solution

The Framework makes AI self-gauge its confidence before starting to work.

Solution

It also makes AI self-gauge the variability of its plan: if there are alternatives with subtle pros and cons, show them to the user.

Problem

AI forgets

Bad fix

You keep repeating yourself (previous decisions, business context, …).

Solution

AI retains knowledge via files such as knowledge-base/DECISIONS.md and knowledge-base/INSIGHTS.md.

Autonomous AI

Focus on what matters, let AI do the rest.

Autonomous Product Management

Let AI autonomously:

✓Test & review
✓Research, plan, and prioritize your tickets
✓Turn your team conversations into tickets
✓Save your team conversations to conversations/<DATE>_<TOPICS_SLUG>.md
✓Market research

Autonomous Coding

Let AI autonomously implement:

✓Quick-wins
✓Quality refactoring
✓UX improvements
✓Consensual work

Note Agents still ask for your confirmation for non-obvious decisions.

Note You're still in control — only pick the autonomicity that works for you.

How it works

The Framework introduces two major building blocks:

1. Enhanced System Prompt

The Framework appends a system prompt instructing AI to follow highly effective practices — such as dividing large work into subtasks, and listing significant alternative solutions.

✓Anti-laziness
✓No important decisions without asking
✓Improved user interaction
✓Improved planning

Note You can customize it, or fully opt-out.

2. Queues

AI Queue

Queue of upcoming AI tasks — tasks are added by humans, or by agents (autonomously if no uncertainty, or after human confirmation otherwise).

Agents autonomously populating the AI queue is what makes AI autonomous.

Technically, it's just a TODO_AGENTS.md file in your Git repositories.

Human Queue

Queue of pending human reviews — populated when agents need you to review (important) decisions with subtle pros and cons.

It's your cockpit — it keeps humans under control.

Note Everything is powered by open source state-of-the-art prompts (you can also bring your own):

Security auditCode qualityResearchProduct Management＋ Add yours

Features

Bring your own subscription

Use your AI subscription — The Framework orchestrates agents via your Claude Code / Codex installation, like humans do.

Optimal quota usage

Stop wasting unused quota — The Framework drives AI autonomy for optimal quota usage: maximum daily usage while keeping enough quota space for manual prompts.