The agent that wrote the code shouldn't be the one that says it works.

Independent executor–verifier orchestration for software changes.

Install

npm install -g @the-open-engine/zeroshot
zeroshot

Requires Node ≥ 22 and one supported provider. Guided setup detects installed providers, chooses a default, and configures worktree isolation for fresh repositories. Linux and macOS today; Windows is deferred.

First run

cd your-repo
zeroshot run "Add a --json flag with tests"

In a git repository, the guided default runs in a separate worktree, so the current checkout is not edited. Use --no-isolation only when you explicitly want the run to modify the current checkout.

Observe the result from another terminal:

zeroshot list
zeroshot logs <id> -f

Unattended, 100× speed · 90-minute run · 5 iterations to approval.

How it works

Zeroshot drives a coding agent through an executor–verifier loop. A conductor sizes the workflow, an executor implements the change in an isolated workspace, and a separate verifier judges the observable result. Validators do not share the executor's session or reasoning context; they may receive explicit handoff artifacts and must reproduce reported failures. The loop continues until the change is verified or returns a concrete reason it is not.

Every step is written to a crash-safe SQLite ledger. Bring your own provider and backend: Zeroshot orchestrates their CLIs without storing provider keys.

One run: classify, execute, verify, and repeat when evidence fails.

Providers and issue sources

Provider engines come from the registry: Claude, Codex, bundled Gateway, Gemini, OpenCode, Pi, OMP, Kiro, and Copilot. Model gateways stay behind the single Gateway provider.

zeroshot providers
zeroshot providers set-default codex
zeroshot run 123 --provider gemini

Issue sources are auto-detected from repository context or explicit URLs: GitHub, GitLab, Jira, Azure DevOps, and Linear. Each source requires its own authenticated client where applicable. See docs/providers.md for provider installation, model levels, and credentials.

Isolation and delivery

Guided setup defaults fresh repositories to git worktree isolation. Delivery flags cascade: --ship implies --pr, which implies --worktree.

Mode	Flag	Behavior
Git worktree	--worktree	isolated branch and checkout; guided default
Docker	--docker	container isolation for riskier workloads
Current checkout	--no-isolation	explicit escape hatch; modifies the active checkout
Pull request	--pr	worktree plus PR creation
Ship	--ship	worktree, PR, and merge after approval

Command reference

zeroshot run <input>             # issue, URL, markdown file, or inline text
zeroshot run 123 --docker        # container isolation
zeroshot run 123 --pr            # worktree + pull request
zeroshot run 123 --ship          # worktree + PR + merge after approval
zeroshot run 123 -d              # background run

zeroshot list                    # tasks and clusters (--json)
zeroshot status <id>             # detailed status (--json)
zeroshot logs <id> -f            # stream logs
zeroshot resume <id> [prompt]    # resume a stopped or failed run
zeroshot stop <id>               # graceful stop
zeroshot kill <id>               # force stop

zeroshot providers               # provider availability and defaults
zeroshot settings                # effective settings
zeroshot agents list             # available agents

Docker credential configuration

Docker mounts and environment forwarding are explicit and provider-aware. Defaults include gh, git, and ssh; provider-specific authentication follows the registry contract.

zeroshot settings set dockerMounts '["gh","git","ssh","aws"]'
zeroshot run 123 --docker --mount ~/.aws:/root/.aws:ro
zeroshot run 123 --docker --no-mounts

See docs/providers.md for details.

Scope and status

Zeroshot performs best when a task has clear acceptance criteria. If you can't say what "done" means, an independent verifier can't confirm it.

Task	Good fit?	Why
Add rate limiting (sliding window, per-IP, 429)	Yes	clear requirements
Refactor auth to JWT	Yes	defined end state
Fix a login bug	Yes	success is measurable
"Make the app faster"	No	needs exploration first
"Improve the codebase"	No	no acceptance criteria

• Pre-1.0 in spirit. Interfaces still move between releases; pin your version. (The npm version auto-increments on every merge, so read it as a build counter, not a stability promise.)
• Crash-safe. All state persists to a SQLite ledger; zeroshot resume <id> continues at any time.
• No TUI in this release. Monitor with zeroshot logs <id> -f, zeroshot list, and zeroshot status <id>.

Architecture, quality gates and command proofs

Zeroshot is a message-driven coordination layer: a conductor classifies each task by complexity and type, a workflow template selects agents and validators, agents publish results to a SQLite ledger, and validators approve or reject with specific findings.

• Required handoff quality gates: in --pr/--ship flows, the git-pusher fails closed until every configured gate has fresh passing evidence.
• Cmdproof: make expensive exact commands reusable across agents with zeroshot cmdproof check <id>.

See CLAUDE.md for the cluster schema, primitives, and the conductor's classification model.

The Open Engine

Zeroshot is Layer 01 · Verification of The Open Engine, the open stack for autonomous software production. Generating code is easy; trusting it is not. The engine is layered because trust is layered:

	Layer	Status
01	Verification: Zeroshot	This repo · open · shipping
02	Constraints: Opcore	Sibling · alpha
03-05	Intent · Context · Runtime	In development

Zeroshot runs the loop: an agent writes the change, and an independent verifier decides whether it holds: approve, or a reproducible failure. Opcore is the sibling layer, a deterministic, local, read-only constraints gate for coding agents. Zeroshot packages Opcore 0.2.1 and uses introduced-change validation so existing repository debt never blocks an otherwise clean change. Verification asks "does this meet the goal?"; constraints ask "is this within tolerance?"

Each layer ships the same way: extracted from the platform we run, then opened. Trust nothing. Verify everything.

Contributing

See CONTRIBUTING.md for development setup, CODE_OF_CONDUCT.md before participating, and SECURITY.md for security reports. More in docs/ and CLAUDE.md.

Questions and help: Discord.

License

MIT. The Open Engine Company.