Alexandre Augusto Arduini Tavares
← Back to selected work
Independent Project

AgileForge: building software with AI agents

A tool for planning agent work and checking its output

Project details

Project context

Independent Project • Open Source

Project Repository

From requirements to reviewed work
Open video
An accepted Specification links an example Task and Story to coding-agent implementation and verification in the target repository.
Illustrative workflow, not a product screenshot. One Project links a human-accepted Specification to planning, coding-agent implementation, verification and recorded evidence. Story closure, Sprint review, closure and learning are separate steps. AgileForge rules choose eligible actions; ADK generates candidates. AI responses still vary and need review. Independent open-source project.

Problem

AI agents can produce code quickly. Keeping that work tied to approved requirements is a separate problem. I built AgileForge to make those requirements, the work in progress and the resulting files easier to inspect.

Architecture

A person approves the product goal and specification before planning and execution can begin. Rules in the application decide where work goes next. Google ADK handles the tasks assigned to AI models.

Outputs follow defined data formats and retain links to their inputs. Content fingerprints and saved versions make it possible to check which requirements an output came from. The workflow uses stored facts to decide whether a step can proceed and where to resume after a failure.

Using the tool

I built a Python CLI and a local dashboard to inspect tasks, requests and decisions, and see the next available action. Automated checks support development of the tool itself.

Independent project

AgileForge is my independent open-source project. Its routing rules control how work moves between steps. AI responses can still vary and need review.

The code and documentation are available in the public repository.

← Back to selected work Contact Alexandre