About this project
Zeroshot is an independent executor–verifier orchestration tool for software changes. Its core premise is stated plainly: the agent that writes the code should not be the one that decides it works. Instead of hiding agent loops inside prompts, Zeroshot makes control flow authored data — sequence, parallel review, retry paths, delivery, and exit conditions are all explicit before a run starts.
The built-in software-change graph gives a goal to a worker, runs acceptance and code review independently and in parallel, routes rejected evidence to a repair worker and repeats both reviews, then (with delivery enabled) delivers an accepted change through Git, CI, and merge, routing delivery conflicts back through repair and review. No runtime agent chooses the next step; every transition is bounded and every event is written to a durable SQLite ledger.
A second built-in template, auto-research, runs ten bounded iterations with explorer, synthesizer, and challenger scouts proposing ways to advance a charter, plus evidence, method, and progress judges reviewing the selected experiment. Unanimous adopt keeps workspace changes; record_only preserves a supported negative or inconclusive result while restoring them; abort rejects invalid or incomplete evidence. State, summaries, backlog, iteration records, and reversible backups live under .zeroshot/research.
Installation is via npm (npm install -g @the-open-engine-company/zeroshot), which installs a verified native binary for Linux x64/arm64, macOS x64/arm64, or Windows x64, plus a Zeroshot skill for Codex, GitHub Copilot, and Claude Code at user scope. Native archives and checksums are attached to each canonical vX.Y.Z GitHub Release. The README notes that v8 is a hard interface cutover: the former Node.js runtime is retired and the native zeroshot executable is the product.
The same graph and runtime plan run against three environments. Local mode runs directly in a Git workspace and can reuse an existing Codex or Claude Code login, including subscription-backed sessions, or bind explicit provider credentials. A browser UI (zeroshot ui, default http://127.0.0.1:4173/ui/) edits profiles and inspects live or completed runs, sharing the CLI's saved profiles and history. An experimental ACP mode exposes a saved local profile as an ACP agent.
Self-hosted mode runs a Docker target image containing the native engine plus pinned Codex, Claude, and GitHub Copilot harness CLIs, keeping execution and durable state on infrastructure you control; the target also serves a profile editor and run viewer. Zeroshot Cloud uses a built-in cloud target for a shared team queue and durable run history, with device-code login. Failed runs with a recoverable workspace can be resumed on the latest saved files or from an earlier node boundary via zeroshot resume and zeroshot checkpoints.
Delivery is explicit: keep work local, push the managed branch, prepare a mergeable PR, or ship it; PR and ship runs process visible GitHub feedback unless their delivery runtime binding opts out. The repository also references a Python SDK, a cluster API reference, and an OpenEngine graph contract. Development uses npm for the static UI, repository tooling, and npm delivery, with Rust serving the UI; checks include npm run check and cargo test --workspace. Licensed under MIT.
Comments
0 Rating appears after 10 ratings
Sign in to join the discussion.