[#999] Add GPT-5.6 Sol/Terra/Luna to codex model dropdown

[realproject7]

Closes #999

EPIC Alignment

• Parent: none — standalone enhancement, NOT part of EPIC [EPIC] QuadWork v2.4.0 hardening & quality pass — audit remediation #967. Standard QuadWork merge flow applies.
• This PR's role: Append the GPT-5.6 Sol/Terra/Luna concrete slugs to the codex model dropdown so operators can select them without hand-editing ~/.quadwork/config.json.
• Contracts consumed/exposed: none changed — pure data addition to MODEL_OPTIONS.codex. effectiveModel/sanitizeModel already accept any concrete slug.
• Fits with merged siblings: same append pattern as the Add Fable 5 (claude-fable-5) to the agent model catalog #958 Fable-5 row; ordering invariant from Settings: agent Model dropdown hardcoded to Claude models — not provider-aware (shows 'sonnet' for Codex/Gemini agents) #931 preserved.

Self-Verification

• Scope: 2 files, +7/−0. src/lib/agentModels.ts appends 3 rows AFTER the existing concrete codex rows (label == value); server/agentModels.test.js adds one assertion.
• Ordering invariant: modelsForBackend("codex")[0] stays "gpt-5.4" (the Settings: agent Model dropdown hardcoded to Claude models — not provider-aware (shows 'sonnet' for Codex/Gemini agents) #931 default/heal target) — no reorder. Existing test line 38 still guards this.
• Behavior-preserving: no logic changes; effectiveModel/sanitizeModel unchanged.
• Tests: node server/run-tests.js → 64 passed, 0 failed, 2 skipped (Jest-style, pre-existing Unref routes.js module-load pollers + add cross-platform 'npm test' (fix full-suite hang) #836 skips). New assertion: optionsForBackend("codex") includes gpt-5.6-sol/gpt-5.6-terra/gpt-5.6-luna → PASS.
• Typecheck: npx tsc --noEmit → clean (exit 0).

Deviations

none

🤖 Generated with Claude Code

[realproject7]

@re2 APPROVE at c29d5ade690e9a75454c8dbd021be676c15ede4c

Pure data addition per #999, behavior-preserving for existing agents. Standard QuadWork flow (standalone, not EPIC #967).

Checked (evidence)

• Scope: gh pr view 1000 → 2 files, +7/−0, base main, MERGEABLE. Only src/lib/agentModels.ts (+6) and server/agentModels.test.js (+1).
• Append, no reorder: src/lib/agentModels.ts:16-18 appends gpt-5.6-sol / gpt-5.6-terra / gpt-5.6-luna (label == value) AFTER the existing concrete rows (gpt-5.4/gpt-5/gpt-4o at lines 13-15). Same pattern as the Add Fable 5 (claude-fable-5) to the agent model catalog #958 Fable-5 append.
• Settings: agent Model dropdown hardcoded to Claude models — not provider-aware (shows 'sonnet' for Codex/Gemini agents) #931 invariant preserved: ran modelsForBackend("codex")[0] at this SHA → gpt-5.4 (the default/heal target). Full codex list order: "",gpt-5.4,gpt-5,gpt-4o,gpt-5.6-sol,gpt-5.6-terra,gpt-5.6-luna.
• No logic change: effectiveModel/sanitizeModel (src/lib/agentModels.ts) already accept any concrete slug via opts.some(o => o.value === model) — new slugs resolve/persist with no code change.
• Test: server/agentModels.test.js:34 new assertion checks optionsForBackend("codex") includes all 3 new slugs. Ran the file locally at this SHA (Node type-stripping requires the real shared module) → 28 passed, 0 failed, including the new Add GPT-5.6 Sol/Terra/Luna to the codex model dropdown #999 case and the existing Settings: agent Model dropdown hardcoded to Claude models — not provider-aware (shows 'sonnet' for Codex/Gemini agents) #931 codex[0] guard.
• CI: gh pr checks 1000 → test pass 51s.

No concerns. Both-approval + CI-green gate met on the review side; @Head merges per standard rule.

[project7-interns]

Verdict: APPROVE

Epic Alignment: PASS

Standalone #999 contract is met: only MODEL_OPTIONS.codex gains the three GPT-5.6 slugs, labels match values, and modelsForBackend("codex")[0] remains gpt-5.4.

Checked (evidence)

• Structural gate: PR body has filled ## EPIC Alignment and ## Self-Verification; UI design table not required for this data-only dropdown option change.
• Issue context: gh issue view 999 → standalone, parent none; contract requires src/lib/agentModels.ts codex-only append, no reorder, no logic/spawn/UI refactors.
• Scope: gh pr diff 1000 / REST files → exactly src/lib/agentModels.ts (+6) and server/agentModels.test.js (+1), no deletions.
• Model rows: src/lib/agentModels.ts:19, src/lib/agentModels.ts:20, src/lib/agentModels.ts:21 add gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna with label == value after existing concrete codex rows.
• Ordering invariant: server/agentModels.test.js:39 still pins modelsForBackend("codex")[0] === "gpt-5.4".
• New coverage: server/agentModels.test.js:34 asserts optionsForBackend("codex") includes all three new slugs.
• Consumer path: src/components/AgentModelsWidget.tsx:25 and src/components/AgentModelsWidget.tsx:248 use optionsForBackend; src/components/SettingsPage.tsx:6 imports shared helpers from @/lib/agentModels.
• Riskiest part of this diff: stale duplicate model catalogs would make the UI miss the new slugs; checked PR head consumers and they read the shared helper, so acceptable.
• Kill-list: scanned all items — clean.
• CI: gh pr checks 1000 → test pass 51s.

Findings

None.

Decision

Approve. The PR is the requested data-only addition with focused test coverage and no behavior-changing logic outside the model catalog.