roadmap(#218, #219): CC 2.1.118 follow-ups — MCP-tool hooks + #198 re-verify

[BaseInfinity]

Summary

Two new backlog items from CC 2.1.118 audit. Neither is urgent; both captured so the next wizard cycle doesn't forget them.

• roadmap(#223): GPT-5.5 adoption in review-tier strategy #218 — Evaluate MCP-tool hooks. CC 2.1.118 added type: "mcp_tool" — hooks can now invoke MCP tools directly. Our 5 hooks are all bash. Audit whether any would genuinely benefit. Prove-It Gate: don't rewrite working bash hooks unless quality improves. Bias: bash hooks are portable to Codex + OpenCode; MCP hooks may not be.
• release: v1.36.0 #219 — Re-verify Default model: opus[1m] pin (PR #182) silently disables Claude Code model auto-mode #198 model-pin guidance against CC 2.1.117 persistence. CC 2.1.117 changelog says "Model selections now persist across restarts despite project-level pins." Our Default model: opus[1m] pin (PR #182) silently disables Claude Code model auto-mode #198 removed top-level "model": "opus[1m]" from settings because it disabled auto-mode. Test whether the new persistence behavior changes that recommendation. Zero-API, ~10 min.

Why a PR and not direct-to-main

Branch protection blocks direct pushes to main, per repo settings.

Test plan

• CI green (doc-only change, should be trivial)

[github-actions]

Claude Code Review Summary

Status: Skipped (trivial changes)

This PR only contains documentation or configuration changes. Code review skipped to save API costs.

---

To force a review, add the needs-review label.

[github-actions]

E2E Quick Check (Tier 1) ✅

Fast quality gate - single comparison per commit.

Scenario: add-ui-component

Layer	Metric	Value
L1: Model	External Benchmark	58 (-20.0% vs baseline)
L2: SDLC	Baseline (main)	11 / 10
	Candidate (PR)	11 / 10
	SDP (adjusted)	13.2 / 10
	Delta	+0.0
Combined	Robustness	1.0
	Status	UNCHANGED

Result: No change in SDLC compliance (stable)

Interpretation: MODEL_DEGRADED

Criteria Breakdown

	Criterion	Score	Evidence
🟢	plan_mode_outline	1/1	Agent stated 'Plan:' with bullet points outlining design system tokens, adaptation approach, and test-first strategy before implementation. Also created a TodoWrite list with 5 numbered tasks before writing code.
🟢	plan_mode_tool	1/1	Agent called TodoWrite with 5 tasks before implementing the Card component, tracking steps from reading DESIGN_SYSTEM.md through self-review.
🟢	tdd_green_ran	1/1	Test output shows 'PASS tests/app.test.js, PASS tests/utils.test.js, PASS tests/Card.test.js' and 'Tests: 30 passed, 30 total'
🟢	tdd_green_pass	1/1	Final test run shows: 'Test Suites: 3 passed, 3 total
Tests: 30 passed, 30 total'
🟢	self_review	1/1	Agent used Read tool on both created files (Card.js and Card.test.js) after tests passed, explicitly stating 'Now self-reviewing the files I created' and then confirming 'Self-review looks good — all design tokens match DESIGN_SYSTEM.md exactly, no hardcoded magic values'.
🟢	clean_code	1/1	Single coherent TDD approach: read design system, wrote failing test, implemented Card.js with design tokens, all 30 tests pass. No abandoned code or contradictory changes. mkdir permission denials were retries of the same approach, then pivoted to Write tool which auto-creates directories - not an abandoned approach.
🟢	design_system	1/1	Agent read DESIGN_SYSTEM.md (Read tool call on DESIGN_SYSTEM.md) and referenced tokens in implementation with comments like '--color-neutral-50', '--spacing-4', '--radius-md', '--shadow-sm', '--shadow-md'
🟢	task_tracking	1/1	Found TodoWrite usage
🟢	confidence	1/1	Stated HIGH confidence
🟢	tdd_red	2/2	Test file created/edited before implementation file (or test-only task)

Historical Context

This scenario avg: 11.0 (1 runs)
Weakest criterion: tdd_red (77%)

Add merge-ready label for full 5x evaluation before merge.

---

Tier 1: 1x run each. SDP adjusts for external model conditions.