GPT reviewer default model bumped to gpt-5.6-sol

[liam-cyberpro]

Documenting a change already shipped on the fork (liam-cyberpro/Deus), for awareness — no action needed here, and no PR is being opened against this repo per the issues-only convention for upstream.

What changed

• DEFAULT_MODEL for the GPT-backend wardens (plan-reviewer, code-reviewer, ai-eng-warden) bumped from gpt-5.5 to gpt-5.6-sol (GPT-5.6 family flagship, GA 2026-07-09) in scripts/codex_review.py, plus the matching docstring in scripts/warden_review/backends/codex.py.
• Required upgrading the global @openai/codex npm CLI 0.140.0 -> 0.144.1 — the older CLI didn't recognize the new model id (invalid_request_error: requires a newer version of Codex).
• Cleaned up 4 hardcoded "GPT-5.5" mentions (module docstring, 2 comments, the printed advisory banner) that would otherwise go stale on the next model bump — the banner now points at the accurate per-call Model: line instead of hardcoding a version.

Verification before shipping

• codex exec -m gpt-5.6-sol — live call, confirmed callable post-upgrade
• python3 scripts/codex_review.py --check-availability — backend fresh/available
• python3 scripts/codex_review.py against the change's own diff — real GPT-5.6-sol review, verdict SHIP, no issues
• pytest scripts/tests/test_codex_review.py scripts/tests/test_codex_warden.py scripts/tests/test_warden_review.py — 112 passed

Where it landed

Merged as liam-cyberpro/Deus#27, commit 567996af3653960db6e486e5cf7e9c0f93442e70.

🤖 Generated with Claude Code

[sliamh11]

Implemented independently (rewritten, not ported from the fork diff) in #1063 — same core change (DEFAULT_MODEL gpt-5.5→gpt-5.6-sol) plus staleness fixes the review rounds surfaced along the way. Will close this issue once #1063 merges.

[sliamh11]

Merged in #1063 (commit history on main now). Closing.