no-human-ai/no_human
From ticket to reviewed pull request. Free and open-source, on your machine.
What it solves
no_human is an AI coding factory that automates the entire software development lifecycle from ticket intake to a reviewed pull request. It aims to eliminate the 'hallucination' and trust issues common in AI coders by implementing strict adversarial reviews, mechanical tamper guards, and mandatory proof-of-fix verification.
How it works
The system operates as a loop that integrates with issue trackers (Jira, Linear, monday.com) to pull tasks. For each task, it follows a rigorous process:
- Planning: It generates a plan and acceptance criteria based on the ticket and repository analysis before writing any code.
- Implementation: An AI coder implements the changes.
- Tamper Guard: The system mechanically checks if tests were deleted, skipped, or weakened without justification.
- Reproduction Gate: For bug fixes, it verifies that the provided tests fail on the old code and pass on the new code.
- Adversarial Review: A separate AI model, which has not seen the coder's transcript, attempts to refute the "done" status using a pass/fail checklist citing specific files and lines.
- Verification: It runs the project's local or CI tests and opens a pull request only after passing these gates.
Who it’s for
Software engineers and teams who want to automate ticket-to-PR workflows while maintaining high confidence in code quality and security through automated adversarial auditing.
Highlights
- Adversarial Review: Uses a second, independent model to audit changes, preventing self-scoring bias.
- Tamper Guard: Prevents AI from "cheating" by deleting tests or turning assertions into tautologies.
- Reproduction Gate: Ensures bug fixes are actually fixing the reported issue by testing against the merge base.
- Tracker Integration: Native syncing with Jira, Linear, and monday.com.
- GitHub Action: Provides the same adversarial reviewer and tamper guard as a PR check for manual contributions.
Related
- Project
- Project
- Project
- Project
- Project