feat: add unified_review MCP prompt + sync tool/prompt counts

Adds the unified_review MCP prompt (invoked as /code-review-graph:unified_review)
that drives the unified-review workflow: graph context -> detect_changes ->
score_review -> findings -> dedupe_findings -> generate_report. READ-ONLY:
every finding waits for a manual fix decision. Takes base and tier
(fast/standard/strict) arguments.

- prompts.py: unified_review_prompt + docstring count 5->6
- main.py: @mcp.prompt() registration
- tests/test_prompts.py: TestUnifiedReviewPrompt + preamble test
- docs: LLM-OPTIMIZED-REFERENCE, COMMANDS, CLAUDE.md, INDEX, architecture,
  README + 4 localized READMEs; tool count 30->31 and prompt count 5->6
  synced (historical ROADMAP/FEATURES version notes left unchanged)

code_review_graph/prompts.py

@@ -1,6 +1,6 @@
 """MCP prompt templates for Code Review Graph.
 
-Provides 5 pre-built prompt workflows, all enforcing token-efficient
+Provides 6 pre-built prompt workflows, all enforcing token-efficient
 detail_level="minimal" first patterns with get_minimal_context entry point.
 
 1. review_changes   - pre-commit review using detect_changes + affected_flows
@@ -8,6 +8,7 @@ detail_level="minimal" first patterns with get_minimal_context entry point.
 3. debug_issue      - guided debugging using search, flow tracing
 4. onboard_developer - new dev orientation using stats, architecture, flows
 5. pre_merge_check  - PR readiness with risk scoring, test gaps, dead code
+6. unified_review   - three-layer review: graph context + scoring + dedupe + report
 """
 
 from __future__ import annotations
@@ -157,3 +158,60 @@ def pre_merge_check_prompt(base: str = "HEAD~1") -> list[Message]:
         "7. Output: GO/NO-GO recommendation with 1-sentence "
         "justification + list of required follow-ups."
     )
+
+
+def unified_review_prompt(
+    base: str = "HEAD~1",
+    tier: str = "standard",
+) -> list[Message]:
+    """Three-layer unified review workflow (READ-ONLY).
+
+    Fuses CRG graph context with the objective scoring metrics
+    (score_review), finding merge (dedupe_findings), and the standalone
+    HTML report (generate_report).  Every finding is presented for a manual
+    fix decision -- this workflow never modifies code.
+
+    Args:
+        base: Git ref to diff against. Default: HEAD~1.
+        tier: Review tier.  "fast" (Layer 1 + blockers only),
+            "standard" (all layers), "strict" (full + per-item
+            confirmation).  Default: standard.
+    """
+    tier_notes = {
+        "fast": (
+            "fast tier: run Layers 1 and the blocker check only; "
+            "skip Layer 2 metrics and the report."
+        ),
+        "strict": (
+            "strict tier: full review; every blocker and major finding "
+            "needs per-item user confirmation before it is recorded."
+        ),
+    }.get(tier, "standard tier: run all layers.")
+    return _user(
+        f"{_TOKEN_EFFICIENCY_PREAMBLE}\n"
+        f"## Unified Review Workflow (base={base}, tier={tier})\n"
+        f"{tier_notes}\n"
+        "**READ-ONLY.** Present every finding for a manual fix decision. "
+        "Never modify code, commit, or push.\n"
+        '1. Call `get_minimal_context(task="unified review")` for the '
+        "risk overview.\n"
+        '2. Call `build_or_update_graph()` to ensure the graph is '
+        "current.\n"
+        '3. Call `detect_changes(detail_level="minimal")` for changed '
+        "files, risk score, test gaps and affected flows.\n"
+        '4. Call `score_review(detail_level="standard")` for the '
+        "objective metrics (sql_risk, exception_coverage, redundancy, "
+        "high-risk density, vulnerability).  Trust the tool grades; "
+        "LLM-judged metrics are in `llm_judged`.\n"
+        "5. Review the changed source (Layer 1 chain decomposition) and "
+        "produce findings with severity (blocker/major/minor), "
+        "confidence (1-10), file:line and a proposed fix.\n"
+        '6. Call `dedupe_findings(findings=<your findings>)` to merge '
+        "by fingerprint, boost multi-source confidence and compute the "
+        "PR quality score.\n"
+        '7. Call `generate_report(review_data=<verdict, tier, scope, '
+        'metrics, merged findings>)` to write code-review-report.html.\n'
+        "8. Output: verdict (✅ PASS / ❌ FAIL), severity counts, each "
+        "issue with confidence + fix, and the manual-review items.  "
+        "Any blocker → verdict ❌ FAIL."
+    )