DEV.co
AI Coding Agents · the-open-engine

zeroshot

Zeroshot is an open-source CLI that runs autonomous multi-agent AI workflows to write, review, test, and iterate on code changes until they pass validation. It orchestrates separate planner, implementer, and validator agents to reduce hallucination and improve correctness in AI-generated code.

Source: GitHub — github.com/the-open-engine/zeroshot
1.6k
GitHub stars
140
Forks
JavaScript
Primary language
MIT
License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

FieldValue
Repositorythe-open-engine/zeroshot
Ownerthe-open-engine
Primary languageJavaScript
LicenseMIT — OSI-approved
Stars1.6k
Forks140
Open issues52
Latest releasev6.2.0 (2026-06-27)
Last updated2026-07-06
Sourcehttps://github.com/the-open-engine/zeroshot

What zeroshot is

A Node.js-based agent orchestration framework that shells out to provider CLIs (Claude Code, Codex, Gemini, OpenCode) and runs isolated validation loops with crash-safe SQLite persistence. Supports GitHub/GitLab/Jira/Azure DevOps issue backends and configurable quality gates; no built-in validation tooling—gates are user-defined or external commands.

Quickstart

Get the zeroshot source

Clone the repository and explore it locally.

terminalbash
git clone https://github.com/the-open-engine/zeroshot.gitcd zeroshot# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Autonomous bug fixing with reproducible validation

Use when bugs have clear acceptance criteria and existing test suites. The multi-agent loop catches regressions the implementer misses because validators run blind checks independently.

High-stakes code refactoring (auth, payments, security)

Refactor auth to JWT or swap database drivers where correctness is non-negotiable. Blind validation and isolated agents reduce the risk of subtle errors in critical paths.

Automated issue triage and implementation backlog reduction

For teams with large open-source backlogs or infra teams standardizing agentic workflows. Crash-safe resume and multi-platform issue support enable batch processing.

Implementation considerations

  • Requires at least one provider CLI installed and authenticated (Claude Code, Codex, Gemini, or OpenCode); no built-in language model—all inference is delegated to external services.
  • Isolation modes (none, git worktree, Docker) must be chosen per task; Docker isolation adds complexity but is needed for sensitive or high-concurrency workflows.
  • Quality gates and validators are user-defined; you must script or wire external commands (repo-quality, CI scripts, linters) into the gate configuration; Zeroshot orchestrates but does not provide these tools.
  • Crash recovery relies on SQLite persistence in `~/.zeroshot/`; ensure write permissions and adequate disk space for large workflows or multi-day runs.
  • Node 18+ required; npm install is global (`-g`); ensure PATH and npm prefix are correctly configured in CI or container environments to avoid agent startup failures.

When to avoid it — and what to weigh

  • Task requires exploration or discovery — If the acceptance criteria are undefined ('make the app faster', 'improve the codebase'), agents will loop without clear exit conditions. Use for well-scoped, measurable tasks only.
  • You need native Windows support — Platform support is Linux and macOS only; Windows native and WSL are deferred. Teams requiring Windows must wait for hardening or use Docker workarounds.
  • Validation tooling does not exist or is incomplete — Zeroshot does not provide built-in test runners or linters. You must have existing CI/test infrastructure that agents can invoke; otherwise, blind validation cannot work.
  • Real-time human feedback or live code review is essential — Zeroshot is designed for offline, autonomous loops. If your workflow requires synchronous human input between iterations, a chat-based assistant is more appropriate.

License & commercial use

MIT License (OSI-approved, permissive). Permits commercial use, modification, and distribution with minimal restrictions; requires only license and copyright notice retention.

MIT is a permissive open-source license that explicitly allows commercial use without restrictions or royalty obligations. No commercial license or proprietary terms are documented. Use is permitted in closed-source products and commercial services as long as the MIT license text and attribution are retained. Recommended to review the LICENSE file in the repository for any additional terms, but MIT itself imposes no commercial use barriers.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

SignalAssessment
MaintenanceActive
DocumentationAdequate
License clarityClear
Deployment complexityModerate
DEV.co fitGood
Assessment confidenceMedium
Security considerations

Code execution risk: agents run arbitrary shell commands in isolated environments (none/worktree/Docker); ensure validation scripts are audited and untrusted input is rejected at the task level. Authentication: provider CLIs and GitHub/GitLab credentials are stored locally; protect `~/.zeroshot/` and token directories. Docker mode adds container escape risk and supply-chain risk if base images are not pinned. No security policy, vulnerability disclosure process, or hardening recommendations are documented in the provided excerpt.

Alternatives to consider

GitHub Copilot for pull requests / OpenAI GPT-4 Code Interpreter

Chat-based or single-agent workflows; simpler setup but no blind validation loop or isolated agents; faster for exploration but higher hallucination risk for critical code.

Devin (Cognition) or similar AI engineer platforms

Closed-source, hosted agent platforms; no self-hosting, no control over isolation or validation logic; but mature, well-supported, and require less setup for small teams.

Custom agent orchestration (LangChain, AutoGPT, crews, etc.)

Full control over agent logic, models, and validation; higher development cost and maintenance burden; suitable for teams with strong AI/ML and DevOps expertise who need deep customization.

Software development agency

Build on zeroshot with DEV.co software developers

Start with a local issue: `npm install -g @the-open-engine/zeroshot && zeroshot run '<task>'`. Requires a provider CLI (Claude Code, Codex, Gemini, OpenCode) and existing validation tooling.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

Related on DEV.co

Explore the category and the services that help you build with it.

zeroshot FAQ

Does Zeroshot come with built-in test runners or linters?
No. Zeroshot is an orchestration framework; it shells out to external provider CLIs and user-defined validation commands. You must integrate existing test, lint, and quality tools via the quality gates configuration.
What happens if an agent fails or the process crashes?
Zeroshot persists all state to SQLite (`~/.zeroshot/`). Resume with `zeroshot resume <run-id>` to pick up from the last completed step; no loss of context or cost re-runs.
Can I use Zeroshot in CI/CD pipelines?
Yes, but requires careful setup: provider CLI auth must be available in the CI environment, issue backend integration must be configured (e.g., GitHub token in secrets), and validation tooling must be installed. Docker isolation mode is recommended for CI to avoid state pollution.
Is Windows supported?
Linux and macOS only. Windows (native and WSL) support is deferred. Teams on Windows must use Docker or wait for future releases.

Work with a software development agency

Adopting zeroshot is usually one piece of a larger software development effort. As a software development agency, DEV.co provides software development services and web development expertise — pairing senior software developers and web developers with your team to design, build, and operate ai coding agents software in production.

Ready to automate code with verified agents?

Start with a local issue: `npm install -g @the-open-engine/zeroshot && zeroshot run '<task>'`. Requires a provider CLI (Claude Code, Codex, Gemini, OpenCode) and existing validation tooling.