Archbro

ArchBro is a human-agent project workspace that turns goals into a living architecture and task map, tracks project reality, and flags when execution drifts from design.

Comment

Archbro

Archbro keeps AI-built software aligned by connecting architecture, tasks, and code evidence—while humans retain control of major decisions.

Inspiration

AI coding agents can build features quickly, but project context often falls behind. Architecture diagrams become outdated, tasks drift away from the original design, and every new agent must reconstruct what the team already knows.

We built Archbro to give humans and AI agents one shared source of truth.

What it does

A user starts with a project goal. Archbro then:

  1. Generates a Living Architecture and execution tasks.
  2. Tracks task progress and project evidence.
  3. Uses connected repository evidence to compare implementation with the approved design.
  4. Detects implementation issues or genuine architecture drift.
  5. Creates a reviewable proposal when a major change is needed.
  6. Updates the architecture and affected tasks only after human approval.

Routine coordination can be automated, but the agent cannot approve its own architecture changes.

How we built it

Archbro uses the Strands Agents SDK for agent reasoning and tool execution.

It combines:

  • FastAPI and PostgreSQL for canonical project state
  • Firebase Authentication for account security
  • WebMCP and MCP tools for structured project and repository access
  • GitHub evidence for implementation verification
  • Deterministic validation before any agent-generated action is saved
  • A human Accept/Reject boundary for consequential changes

Challenges

The hardest challenge was making the agent useful without giving it unchecked authority. We also had to distinguish normal implementation problems from true architecture drift and ensure external repository content remained evidence—not trusted instructions.

What we learned

Professional agents need more than strong models. They need reliable context, constrained tools, durable state, and clear human decision boundaries.

Archbro demonstrates how AI agents can perform meaningful project coordination while humans remain responsible for the decisions that shape the system.

Built With

Share this project:

Updates

Submission history