agentic-toolbelt

Agentic Toolbelt

Author: Roberto Del Prete

agentic-toolbelt packages a reproducible baseline for agent-driven projects. It ships a curated set of agent instructions, Codex runtime templates, and supporting assets, then installs them into a target repository through a single CLI.

pip install gives you the package. agentic-set applies the payload to the current project directory.

Built for teams that want a consistent agent baseline across repositories without hand-assembling boilerplate.

What It Installs

Running agentic-set copies the packaged baseline into the directory where you launch it:

• .agents
• .codex
• .github
• .tasks
• templates
• AGENTS.md

It also materializes the active Codex runtime:

• .codex/config.toml
• .codex/agents/*.toml

Each file in .codex/agents/*.toml is a standalone Codex custom agent definition with its own name, description, and developer_instructions.

The copied templates/codex/ directory also includes package reference material such as routing-profiles.toml, which documents the shipped orchestration policy without extending the active Codex runtime config.

The bootstrap also creates a root .tasks/ workspace. Use the packaged task-breakdown skill when a request is too large or ambiguous to plan in one pass; it narrows the problem into numbered JSON briefs such as .tasks/001-auth-boundary.json.

Packaged repo skills include plan, task-breakdown, implement, docs, publish, phidown, ssh-transfer, ssh-command, hpc-job, remote-tmux, rsync-dataset, huggingface-datasets, s3-rclone, apptainer, and geospatial-raster. Running agentic-set installs them under .agents/skills/ in the target repository.

The result is a project that can resolve named agents such as planner and implementer immediately after bootstrap.

Install

agentic-toolbelt supports Python 3.9 and newer.

From the repository root:

python -m pip install .

Then bootstrap the current project:

agentic-set

For local development and test runs:

python -m pip install -e ".[test]"

That registers the agentic-set console script in your environment. It does not copy files into a project until you run the command.

Usage

From any project directory:

agentic-set

By default, existing destination paths are preserved. Use --force to overwrite .agents, .codex, .github, .tasks, templates, and AGENTS.md when you need to refresh a project.

agentic-set --force

The CLI uses temporary staging during copy operations so installs are predictable and partial writes are avoided.

Why It Matters

This repository standardizes how autonomous agents are configured, launched, and governed across codebases. The goal is a consistent baseline that is easy to bootstrap, easy to inspect, and straightforward to extend.

Docs

A visual overview is available at:

• docs/index.html

Open it in a browser to review the repo concept, architecture, and CLI workflows.

Development

• The Python package entrypoint is declared in both setup.py and pyproject.toml.
• The current package version is maintained in both files as well.

Release Checks

Build and verify the distributable artifacts from the repository root:

python -m pytest
python -m build
python -m twine check dist/*

For a smoke test of the built wheel, install it into a clean environment and run:

agentic-set --help