Summary

This repository, betta-tech/ejemplo-harness-subagentes, serves as a practical demonstration of Harness Engineering principles applied to AI agent orchestration. It tackles the challenge of enabling autonomous and verifiable AI development by structuring a minimalist Python CLI notes application. The project targets engineers and researchers seeking to understand and implement robust AI workflows, offering a clear blueprint for AI agents to collaborate on software development tasks with auditable progress.

How It Works

The core architecture revolves around a multi-agent system comprising a leader, implementer, and reviewer, orchestrated via a "repository-as-the-system" paradigm. Agents interact indirectly through persistent storage (files on disk) rather than direct chat communication, mitigating information loss and enabling step-by-step auditing. The leader defines the plan, delegates tasks to the implementer, and the reviewer validates against predefined criteria. This "anti-telephone-descompuesto" pattern ensures that code changes and test outputs are explicitly recorded in files like progress/impl_<feature>.md and feature_list.json, maintaining a verifiable trace.

Quick Start & Requirements

To initialize the harness and verify its setup, run: bash ./init.sh This script performs essential checks and setup. The primary interaction with AI agents is intended via Claude Code, by opening it at the repo root and issuing commands like "implement the next pending feature." The application itself can be run using Python 3: python3 -m src.cli add "comprar pan" --body "y leche" python3 -m src.cli list

Highlighted Details

• Progressive Disclosure: Agent instructions are managed via AGENTS.md, providing a map for agents to seek information on demand rather than receiving a monolithic prompt.
• Atomic Feature Development: The system enforces a "one feature at a time" workflow, validated by init.sh and tracked in feature_list.json.
• Persistent State: All agent progress, plans, and outputs are stored on disk within the progress/ directory, ensuring state survives restarts and context window limitations.
• Executable Verification: init.sh runs actual application tests, providing an objective measure of correctness independent of agent self-reporting.

Maintenance & Community

The provided README does not contain specific details regarding notable contributors, sponsorships, partnerships, community channels (like Discord/Slack), or a public roadmap.

Licensing & Compatibility

The README does not specify a software license. Consequently, compatibility for commercial use or linking with closed-source projects cannot be determined from the provided information.

Limitations & Caveats

This repository is presented as an illustrative example ("ejemplo-harness") demonstrating AI agent workflow principles. The application code (a minimalist notes CLI) is intentionally simple, with the focus being on the harness structure rather than application complexity. No specific limitations regarding unsupported platforms, alpha status, or known bugs are detailed.