Qorgan42

Inspiration

I built Qorgan42 because I kept running into the same uncomfortable moment with coding agents: the agent says the task is done, the tests passed, and everything sounds fine — but I still do not know exactly what changed.

Did it touch only the files I expected? Did it leave temporary state behind? Did it change helper scripts, test commands, release paths, or Git identity? Qorgan42 started from that gap between “the agent says it is done” and “I can safely accept this into my project.”

The rule behind Qorgan42 is simple: unknown evidence should not silently become success.

What it does

Qorgan42 is an Apache-2.0-licensed, local-first core for safer AI-assisted software work. It acts as a workflow barrier between coding agents and a protected repository.

Instead of accepting an agent's claim that work is done, Qorgan42 makes the handoff explicit:

isolated task worktrees
human-approved read and write scope
exact approved test commands
dependency-chain review
exact candidate approval
controlled code, plan, and release promotion

If required facts are missing, unsafe, or outside the approved boundary, the operation is blocked or surfaced for human review.

How we built it

During OpenAI Build Week 2026, Qorgan42 was extended through exactly 13 implementation commits for Tasks 10-22. The work covered workspace/schema hygiene, strict Project Map authority, guided initialization and repair, task isolation, task planning and scope approval, approved test execution, dependency-chain review, exact approval, controlled code promotion, controlled plan promotion, and controlled release operation.

The submitted repository preserves the main branch history required for judges to verify those commits. The submitted commit is:

2e404699e34529384edc959c0839a83037ab6242

How we used Codex and GPT-5.6

GPT-5.6 was used in the primary Codex build thread to implement, test, review, and document Qorgan42's protected task lifecycle, verification pipeline, approval controls, and controlled promotion and release workflows.

Qorgan42 also dogfooded its own model: a tool built to govern coding agents was itself built with Codex and checked through Qorgan42's own handoff discipline.

/feedback Session ID:

019f4fd7-cade-77f0-872c-daa9358dea75

Challenges we ran into

The main challenge was keeping the submission honest: preserving the Git history needed for evidence while not exposing unrelated sandbox branches, tags, legacy refs, or work-in-progress planning files.

Accomplishments that we're proud of

Qorgan42 reached a coherent protected workflow: the same candidate patch, tree, test evidence, Git identity, and approval record move through review before promotion.