Back to directory

no_human

A free, open-source coding pipeline that turns a ticket into a reviewed pull request on your machine: it plans before writing code, has a second model adversarially review the result, and will not claim a fix works unless its tests fail on the old code and pass on the new one.

no_human

no_human runs a full ticket-to-pull-request loop locally: it drafts a plan from the ticket and what it finds in the repository before writing code, and skips planning only for changes it judges trivial, always recording that decision in the run's event stream. A separate model, in a session that never saw the coder's own transcript, is then told to refute that the work is done, producing a pass or fail checklist that cites specific files and lines rather than a numeric self-score.

Before that review runs, a tamper guard mechanically checks for deleted tests, new test skips, or an assertion turned into a tautology, and any of those stops the attempt outright. For bug fixes, the tests offered as proof must fail at the code's previous state and pass on the new one, which the project calls its reproduction gate; other change types can be bound to the same gate through configuration. When the agent cannot finish, it stops and states why, either a specific blocking question or a structured record of running out of budget, rather than producing an invented diff.

Over 65 days building itself, the project reports several hundred of its own claimed-done attempts were sent back by this review process. It installs with a package-manager tool install command, and also ships a desktop app.

More in Coding agents

See all 793
  • The City

    github.com

    A native macOS app that shows Claude Code runs as a city, with each project as a building, each team as a floor of robots, and every animation driven by a real event from the claude CLI's stream.

  • rig

    tryrig.ai

    A small, single-binary agent runtime that points at any OpenAI-compatible model, local or hosted, reads and edits a repository's code, runs its tests and remembers what it learned, with a loop implemented in a few hundred lines.

  • Jauvex

    jauvex.reindent.com

    An Electron desktop app that puts Claude Code, Codex and Grok coding-agent sessions side by side and lets you steer them by voice, with a small model handling fast routing decisions.