No_human open-sources a local AI workflow from ticket to reviewed pull request

Summary

No_human is a free, open-source tool that runs an AI coding workflow on the user's machine, taking a ticket through planning, implementation, testing, independent review, and pull-request creation. It creates a plan from the ticket and repository, although it can skip planning for changes it judges trivial and records that decision in the event stream. A second model reviews the work in a session that does not see the coder's transcript, producing a pass/fail checklist with file and line citations rather than a numeric self-score. Before review, a tamper guard mechanically checks for deleted tests, new skips, and tautological assertions. Its reproduction gate requires proposed evidence tests to fail at the merge base and pass on the new tree; by default this applies to Python bug fixes, while a configuration option can require it for every change. Tests can run locally or through optional CI, and missing test commands are shown as NOT RUN. When the system cannot finish, it stops with a specific question or a structured budget record instead of inventing a diff. The project provides a CLI, local board, desktop releases with SHA-256 files, tracker integrations for Jira, Linear, and monday.com, issue imports from GitHub and GitLab, and CI integrations with Jenkins and CircleCI. Its MCP server exposes task creation and status tools to clients such as Claude Code and Cursor through a localhost stdio bridge. Installation requires Python 3.12 or newer, uv, git, and Node/npm when building the board from source; the project is released under the MIT license.