the-open-engine/zeroshot

Independent executor–verifier orchestration for software changes.

What it solves

Zeroshot addresses the lack of trust and reproducibility in AI-driven software engineering. It prevents the "hallucination" of success by ensuring that the agent writing the code is not the same agent that verifies it, separating implementation from verification.

How it works

The system turns a software goal into an explicit multi-agent graph. Instead of relying on hidden agent loops within prompts, Zeroshot uses authored data to define the control flow—including sequences, parallel reviews, and retry paths.

In its standard software-change graph, the process follows these steps:

  1. A worker agent implements the goal.
  2. Independent agents perform acceptance and code reviews in parallel.
  3. If reviews fail, the evidence is routed back to a repair worker for correction.
  4. Once checks pass, the change is delivered via Git, CI, and merge.
  5. Any delivery conflicts are routed back through the repair and review cycle.

All transitions are bounded and every event is recorded in a durable SQLite ledger.

Who it’s for

Software engineers and teams who want a verifiable, structured approach to AI-assisted coding where the process is governed by a predefined graph rather than unpredictable agentic behavior.

Highlights

  • Graph-Based Orchestration: Control flow is explicit and authored, making it easier to inspect and reproduce.
  • Separation of Concerns: Independent agents handle implementation, review, and repair to ensure higher quality.
  • Flexible Deployment: Supports local execution (using existing Codex or Claude Code logins), self-hosted Docker targets, or a managed cloud environment.
  • Durable State: Uses a SQLite ledger to track every event in the execution graph.

Related

  • Project
    adshao/flounder

    Flounder is an open‑source framework that lets LLM coding agents run autonomous, sandboxed security audits. Given a repo, contract address, transaction hash, or similar clue, it prepares a workspace, maps the attack surface, lets the model write and execute PoC tests in an OCI sandbox, verifies findings locally, optionally confirms them against a real target, and generates markdown reports. It supports many scenarios (blind audits, incident investigations, bug‑bounty work) and works with various model providers via a daemon‑based control plane.

  • Project
    Intrect-io/OpenSwarm

    An autonomous AI agent orchestrator that manages a pipeline of AI code workers to resolve issues from Linear or local trackers and manage pull requests.

  • Project
    ifixai-ai/iFixAi

    An independent auditing framework for AI agents that evaluates operational assurance and business KPI alignment through a graded scorecard system.

  • Project
    PhyAgentOS/PhyAgentOS-core

    An agent framework for embodied AI that uses a decoupled execution gateway and evidence-based verification to ensure robots achieve their goals and recursively improve from experience.

  • Project
    qinshihu/itops-agent-platform

    An enterprise-grade AIOps platform that uses multi-agent AI collaboration to automate the full cycle of alert diagnosis, human-approved repair, and verification.