[docs] Start ues CLAUDE rules.

.claude/rules/commands.md

@@ -0,0 +1,26 @@
+# Commands
+
+## Installation
+
+```bash
+pip install -e .
+```
+
+## Running
+
+```bash
+# Run example
+python example.py
+
+# Run benchmarks
+python bench.py                    # Standard benchmark
+python bench_offload.py            # CPU offload benchmark
+```
+
+## Config Defaults
+
+- `max_num_batched_tokens`: 16384
+- `max_num_seqs`: 512
+- `kvcache_block_size`: 256
+- `gpu_memory_utilization`: 0.9
+- `enforce_eager`: False (enables CUDA graphs)

.claude/rules/testing.md

@@ -0,0 +1,20 @@
+# Testing
+
+## Chunked Attention Test
+
+```bash
+CUDA_VISIBLE_DEVICES=4,5 python tests/test_chunked_attention.py 6 2048 64 2
+# Args: num_gpu_blocks input_len output_len num_prefetch_blocks
+```
+
+## CPU Offload Testing
+
+```bash
+# Basic test with limited GPU blocks to trigger offload
+CUDA_VISIBLE_DEVICES=4,5 python tests/test_chunked_attention.py 6 2048 64 2
+
+# Verify consistency (run multiple times, output should be identical)
+for i in 1 2 3; do
+  CUDA_VISIBLE_DEVICES=4,5 python tests/test_chunked_attention.py 6 2048 32 2 2>&1 | tail -3
+done
+```