[zzxwill]

The Claude Code plugin could inherit Codex-only CLI settings and fail without a reviewer. /codejury:review <pr-url> now always selects Claude as both judge and jury in separate sessions, replacing supplied --judge, --reviewer, --jury, and legacy --agents flags for new and resumed runs. Saved settings are unchanged; reviewers remain read-only and the judge handles fixes.

If an older CLI rejects the same-agent pairing, the skill retries once with the repository CLI using the identical normalized arguments. PR URLs remain required, with the web console enabled and pushes disabled by default. Rename the legacy manual skill from multi-agent-code-review to codejury and update its guidance to use Claude for both roles.

Merged current master and resolved the changelog conflict, preserving the Claude Code and Codex desktop plugins and both test suites.

Validation:

• All 265 tests passed after the merge, including a Claude integration fixture that checks role permissions, separate reviewer sessions, reply routing, and resume behavior.
• Installed-package checks and the Codex plugin ZIP build passed. Claude plugin validation passed with the existing omitted-version and optional-icon warnings; git diff --check passed.
• Real Claude Code invocations exercised default roles, conflicting role flags/aliases, and a saved Codex-judge resume. Each selected only Claude, recovered from the installed 0.8.1 rejection with one repository-CLI retry, and left saved settings unchanged. These used CodeJury --dry-run: they verify command routing and saved role selection, not a real AI review of the target PR.

Follow-up to #116.