Take AI Control Back
A template repository that governs AI coding agent sessions with a tailored AGENTS.md, curated skills, and an isolated Docker/VS Code dev container.
At a Glance
Template repository with a tailored AGENTS.md and optional isolated dev container, released under the MIT License.
Engagement
Available On
Listed Nov 2026
About Take AI Control Back
Take AI Control Back is an MIT-licensed template repository by Giovanni Giorgi (daitangio) for keeping AI coding chats under tighter control. It combines a tailored AGENTS.md, a set of shared agent skills, and an optional ready-to-use dev container for isolation. The README says it is tested on Linux and macOS.
What It Is
The project is a shell-script-based scaffold that you add to your own repository. It leverages the pi.dev coding agent and isolates it with Docker and VS Code dev containers. Two install scripts are provided: copyMe.sh copies the basic features into your project, and cloneAndLink.sh installs the template as a git submodule so it can stay up to date.
How the Setup Works
You define a .devcontainer/devcontainer.env file with your API keys, then run in one of three ways: a VS Code dev container, the bin/runInContainer.sh launcher for a throwaway container on the command line, or without a container (the README advises against using pi.dev that way, since it never asks for command confirmation). The pi.dev configuration, models.json, and sessions are kept in the var directory, which is not meant to be versioned. Pi extensions installed in the container are stored in var/pi-agent.
Liquid Models and Skills
The README describes the idea of liquid models: switching frontier models or providers without changing workflow. Skills under .agents are designed to work across codex, Claude, Copilot, and pi.dev. AGENTS.md documents use of the rtk tool, CLAUDE.md is symlinked to it, and a mark-it-down converter turns documents into a more compact form to reduce tokens.
SDLC Experiments
Demo branches show applications built with AI, including demo/nello-openspec, a Trello clone based on OpenSpec. Feature branches ai-sdlc-copilot, ai-sdlc-pi, and ai-sdlc-claude follow the AWS aidlc-workflows approach; the author notes they are slow and token-costly.
Community Discussions
Be the first to start a conversation about Take AI Control Back
Share your experience with Take AI Control Back, ask questions, or help others learn from your insights.
Pricing
Open Source (MIT)
Template repository with a tailored AGENTS.md and optional isolated dev container, released under the MIT License.
- MIT License (OSI-approved)
- Tailored AGENTS.md template
- Optional Docker/VS Code dev container for isolation
- Provided agent skills compatible with codex, claude and copilot
- Users supply their own model/API keys (external cost)
Capabilities
Key Features
- Tailored AGENTS.md for governing AI chats
- Optional Docker and VS Code dev container for isolation
- Zero-lock-in launcher script bin/runInContainer.sh
- Agent skills compatible with codex, Claude, and Copilot
- Liquid models: switch models and providers without changing workflow
- pi.dev configuration and sessions kept in the var directory
- mark-it-down document converter skill to reduce tokens
- Install via copyMe.sh or cloneAndLink.sh git submodule
- Demo branches including OpenSpec-based nello and AI-SDLC variants
