feat(agent): add CSV benchmark recorder + summary/embedding test scripts and skills

New files:
- scripts/model_benchmark.py — shared CSV recorder for all model tests
- scripts/test_summary_model.py — summary model quality benchmark (coherence, coverage, brevity, hallucination)
- scripts/test_embed_model.py — embedding model benchmark (dimension, cosine accuracy, speed)
- .agents/skills/test-summary-model/SKILL.md — skill for testing summary models
- .agents/skills/test-embed-model/SKILL.md — skill for testing embedding models
- benchmarks/README.md — schema documentation

Updated:
- .agents/skills/test-chat-model/SKILL.md — now also records to CSV

All three scripts write to benchmarks/model_benchmarks.csv with one row
per run per check. The CSV accumulates results across runs for comparison.

scripts/model_benchmark.py

@@ -0,0 +1,132 @@
+"""Shared CSV benchmark recorder for model testing scripts.
+
+All model-test scripts (chat, summary, embedding) write their results
+to ``benchmarks/model_benchmarks.csv`` so results accumulate across
+runs and can be compared with a spreadsheet or a simple query.
+
+CSV columns (one row per *run*, not per turn):
+
+    date,script,model,mode,gate_status,turns,answered,caps,
+    tool_turns,emitted,executed,contract,wall_s,
+    extra_col1,extra_val1,extra_col2,extra_val2
+
+* ``script`` — ``chat`` | ``summary`` | ``embed``
+* ``model`` — the model name (e.g. ``lite``, ``turbo``)
+* ``mode`` — ``fixture`` | ``derived`` | ``quality`` | ``dimension`` |
+  ``cosine`` | ``speed``
+* ``gate_status`` — ``PASS`` | ``FAIL``
+* ``turns`` — number of turns/questions
+* ``answered`` — turns that produced an answer (no LLMError)
+* ``caps`` — turns that hit the round cap
+* ``tool_turns`` — turns that emitted at least one tool call (chat only)
+* ``emitted`` / ``executed`` — tool-call counts (chat only)
+* ``contract`` — well-formed / total calls (chat) or quality score 0-100
+  (summary) or cosine accuracy 0-100 (embed)
+* ``wall_s`` — total wall seconds for the run
+* ``extra_*`` — script-specific secondary metrics (e.g. summary
+  coherence, embed dimension, embed speed per vector)
+
+Usage from any test script::
+
+    from scripts.model_benchmark import bench_write
+
+    bench_write(
+        script="chat",       # or "summary" / "embed"
+        model="lite",
+        mode="fixture",
+        gate_status="PASS",
+        turns=10,
+        answered=10,
+        caps=0,
+        tool_turns=9,
+        emitted=9,
+        executed=9,
+        contract=11,  # numerator (denominator = emitted)
+        wall_s=40.5,
+        # optional extras:
+        contract_denom=12,
+        executed_denom=12,
+    )
+"""
+from __future__ import annotations
+
+import csv
+import os
+from datetime import date
+from pathlib import Path
+
+CSV_PATH = Path(__file__).resolve().parent.parent / "benchmarks" / "model_benchmarks.csv"
+_HEADER = [
+    "date", "script", "model", "mode", "gate_status",
+    "turns", "answered", "caps",
+    "tool_turns", "emitted", "executed",
+    "contract", "wall_s",
+    "contract_denom", "executed_denom",
+    "tool_turns_denom", "extra1_col", "extra1_val",
+    "extra2_col", "extra2_val",
+]
+
+
+def bench_write(
+    *,
+    script: str,
+    model: str,
+    mode: str,
+    gate_status: str,
+    turns: int,
+    answered: int,
+    caps: int,
+    wall_s: float,
+    # chat-specific
+    tool_turns: int = 0,
+    emitted: int = 0,
+    executed: int = 0,
+    contract: int = 0,
+    # denominators (for percentages)
+    contract_denom: int | None = None,
+    executed_denom: int | None = None,
+    tool_turns_denom: int | None = None,
+    # extras (any script)
+    extra1_col: str = "",
+    extra1_val: str = "",
+    extra2_col: str = "",
+    extra2_val: str = "",
+) -> None:
+    """Append one row to the benchmark CSV."""
+    CSV_PATH.parent.mkdir(parents=True, exist_ok=True)
+
+    # Defaults: if denom not given, use the numerator
+    c_denom = contract_denom if contract_denom is not None else (contract if contract > 0 else 0)
+    e_denom = executed_denom if executed_denom is not None else (emitted if emitted > 0 else 0)
+    t_denom = tool_turns_denom if tool_turns_denom is not None else turns
+
+    row = {
+        "date": str(date.today()),
+        "script": script,
+        "model": model,
+        "mode": mode,
+        "gate_status": gate_status,
+        "turns": turns,
+        "answered": answered,
+        "caps": caps,
+        "tool_turns": tool_turns,
+        "tool_turns_denom": t_denom,
+        "emitted": emitted,
+        "executed": executed,
+        "executed_denom": e_denom,
+        "contract": contract,
+        "contract_denom": c_denom,
+        "wall_s": f"{wall_s:.1f}",
+        "extra1_col": extra1_col,
+        "extra1_val": extra1_val,
+        "extra2_col": extra2_col,
+        "extra2_val": extra2_val,
+    }
+
+    file_exists = CSV_PATH.exists() and CSV_PATH.stat().st_size > 0
+
+    with open(CSV_PATH, "a", newline="") as f:
+        writer = csv.DictWriter(f, fieldnames=_HEADER)
+        if not file_exists:
+            writer.writeheader()
+        writer.writerow(row)