# Retired Helix Qwen PR Review

- Retired: 2026-08-18 by owner decision
- Former required check: Qwen Review / Required
- Removed workflows: .github/workflows/deepseek-review.yml and .github/workflows/deepseek-exact-main-handoff.yml
- Removed implementation: all dedicated review adapters, GitHub publisher/controller,
  prompt/schema, and review-only test suites

## Current state

Helix has no repository-triggered AI code review workflow, review implementation,
or review-driven automatic merge workflow. New PRs do not call a model, publish
Qwen Review / Required, or invoke the amber-codex-automation merge controller.
The former Python/Node adapters, GitHub publisher/controller, prompt/schema,
review-only tests and `go-ci` review-adapter invocation have been removed. The
main ruleset has no required-status-check rule, so this removal does not strand
PRs behind a missing Qwen check.

The dedicated ECS runner, its service configuration and its runner-local
provider credential were deliberately not changed. They are infrastructure
state, not an active repository workflow. Do not use the remaining runner as
an implicit review service.

## Historical evidence

The abandoned OpenCode implementation was not adopted as a reliable runtime:
its event transport could complete read-only work without a result-shaped
verdict and its session API could leave a listener after cancellation. A later
small direct Qwen function-calling host proved a six-turn, anchored BLOCK
fixture and a clean cancellation on the isolated runner. That experiment is
historical only; it was not published as a Helix workflow.

## Reintroduction boundary

Any future AI review programme requires a new owner decision and a new design.
It must explicitly define review policy, exact-head binding, credentials,
runner isolation, check identity, merge authority, ruleset effects and fresh
acceptance. Do not restore these deleted workflow files or flip a repository
variable to reactivate the former path.